Converting thermal analysis information

into microstructure information

D.A. Sparkman
MeltLab Systems, Winchester, Virginia

Copyright 2010 David Allen Sparkman

With Iron, vermicular or nodularity checks are performed
ABSTRACT in short times, and recently, ultrasonic nodularity checks
on iron samples have been introduced3. This is an im-
Thermal Analysis has been used in the past to determine provement over the older ultrasonic testing on castings
phase diagrams, which describe temperatures and ener- after the molding, cleaning and fettling processes because
gies produced from crystallization or the reversed melting it can lead to faster corrective actions. Efforts have also
of various crystal substances. Physics describes how ex- been made to remove the human element and standardize
othermic and endothermic reactions relate to microstruc- these results by using a computer to make multiple read-
ture. As the hardware and software for thermal analysis ings under a microscope and average the results. Car-
improve, it can be expected that the refinements will al- bides in iron are usually detected by more careful polish-
low better understanding of casting solidification and mi- ing of a micro in the lab. Shrinkage is usually found after
crostructure. Currently the standard consists of extracting the fact, leading to costly scraping of entire lots of cast-
a microstructure sample from the casting or creating a test ings or ultrasonic or x-ray tests to find the shrink
coupon in the mold or an external mold, and examining it
under the microscope. If thermal analysis could be im- Still, these tests are very time consuming and labor inten-
proved and made more sensitive, then the speed of analy- sive and there can be problems in isolating bad castings
sis and the manpower associated with microstructure because of the normal delays involved in testing. This
analysis could be greatly reduced. This paper examines paper looks at the problems involved in improving ther-
ways of improving that sensitivity and the results from mal analysis and how far it might be able to go in pro-
those improvements in finding carbides, shrinkage, and viding a quicker feedback on casting quality.
oxides in iron, beta crystals, gas, copper, and magnesium
silicide in aluminum, and phosphorus in copper. This paper presents a new way of looking at microstruc-
ture through thermal analysis. The validity of this method
INTRODUCTION starts with the development of a smoothing algorithm to
see beyond the noise that has stymied previous research-
Thermal analysis has been used for some time to measure ers. Once the validity of the smoothing method is ac-
iron flowability (carbon equilivant) and more recently to cepted, then one can move beyond the data preparation
measure carbon, silicon and solidus. Work has been done method, and begin interpreting the meaning of the small
to document calculating chill/inoculation, ferrite/pearlite, arrests. In addition, the importance of the 4th derivative as
magnesium, and shrinkage potential. In aluminum, ther- a means of detecting and qualifying the small arrests will
mal analysis is used to measure the degree of inoculation be demonstrated.
and the degree of modification. In addition, grain size,
percent silicon, freezing range, and the coherency point A lot of the information on the small arrests presented in
have been measured. Dendritic arm spacing has been stu- this paper is new and may be questioned by those unfami-
died1, as well as inoculation of hyper-eutectic aluminums liar with the technology. It is hoped that in the next year,
by phosphorus, using thermal analysis.2 some understanding of these arrests can become part of
the overall knowledge of foundrymen, and the appropriate
These have been documented in many other papers and terminology be developed for using this information in
are common knowledge in the foundry industry. But there production.
are many other features on a thermal analysis curve that
can also benefit the production foundry. The scope of the paper is necessarily limited to introduc-
ing the method, showing the new small arrests that can be
The past problem with thermal analysis practiced as a detected, and looking at what these tools may lead to in
manufacturing tool in the foundry was that it was not sen- the future. The application and correlation of this method
sitive enough to pick up things like pearlite, carbides, for more advanced topics such as flake growth, nodulari-
nodularity, nodule count, and vermicular in iron, or beta ty, aluminum modification, physical testing and other
crystals, gas and grain boundary precipitates in aluminum. advanced applications will need to wait for further testing
So a time consuming process of creating a “micro” be- and for another year.
came a standard operation in the foundries.
DESIGN OF EXPERIMENT the 5 volt DC power supply to filter. A small 0.05 micro-
farad capacitor was added to the circuit to remove this AC
Signal Quality component, based on the load being drawn by the elec-
tronics (figure 2).
The first place to start was improving the signal to noise
ratio. Just like a microscope may have a 1000x setting, if
the sample is not well prepared and flat, that magnifica-
tion is considered “empty”4. So too, before trying to cap-
ture good thermal analysis information, it is necessary to
eliminate external noise and magnify the thermal couple
signal. Foundries radiate electromagnetic energy from
furnaces, from power supplies, powered rails, transfor-
mers, and from electromagnets used to charge furnaces.
This energy is easily converted into voltage in the thermal
couple wire, like a radio antenna, picks up electromagnet-
ic signals and converts them into voltage for the radio
electronics5.
Fig 2. An AC dampening capacitor is added to the trans-
former to filter out AC from high frequency furnaces .

Measuring Electronics
Inexpensive instruments may multiplex signals by com-
bining the grounds of all thermal couples. This proves
Fig 1. A schematic showing the type of shielded K thermal troublesome in two aspects. First, the antenna noise of all
couple wire used in the experiments. the grounds are added together, multiplying their effect on
the instrument and second, individual calibrations are not
The first thing done was to shield the thermal couple wire truly separate. If there is a pickup from a furnace on one
using a commercial wire manufactured with an aluminum line, it can affect all lines, and is therefore hard to isolate.
foil shield and a ground wire (figure 1). The thermal A two wire system, where each thermal couple is wired
couple wire is manufactured with a twist to minimize separately to its own A/D converter, provided more accu-
electromagnetic pickup. The ground wire was grounded to racy7. So in these experiments, only two wire connec-
the housing for the electronics, but not to the sampling tions were used for all measurements.
stand. Especially in iron melting, connecting both ends of
the ground wire would have set up a circuit with the fur- Cold junction sensors are the modern substitutes for an
naces if the stand was placed on any of the metal of a ice water junction8, and are an important part of the sys-
coreless furnace. The wire leads were kept short or in tem. This allows the electronics to cope with the changing
conduit when close to the furnaces and power supplies. temperatures of the electronics.

In one foundry, the unshielded thermal couple wire run The voltage signal is converted into a digital number by
was about 300 feet long, and the thermal analysis instru- an electronic circuit called an A/D converter or an analog
ment was picking up a 10 to 20 degree spike every time to digital converter. These come in various sensitivities
an overhead crane threw a spark when changing powered and speeds. The sensitivity is controlled by the precision
rail segments. The electromagnetic radiation from the of the circuit. Common precisions are 12, 16, and 24 bit.
spark was simply being picked up by a rather long 300 Maximum range on the commonly used k-type thermal
foot antenna. Replacing the wire with shielded wire fixed couple in theory is 2540 F (1393 C) at which point the
the problem. thermal couple melts. Practical maximum range is closer
to 2500 F (1371 C) degrees due to softening of the ther-
Another source of electrical noise comes in through the mal couple joint and slight variations in the chemistry of
electrical power. With all sensitive instruments, it is ne- the thermal couple. This 2500 F would translate into
cessary to have a good ground as a reference to measure 54.84875 millivolts.
the amplitude of the signal6. Large differences in voltage
between the ground and the neutral lines suggest a noisy For example, if the measurement was of a signal of 50
electrical source, so that was checked and minimized as millivolts (0.0500 volts or about 2250.24 degrees F), the
well. A final electrical “contamination” could be found 12 bit converter would have a precision determined by its
from high frequency furnaces that superimposed an AC least significant bit varying between a 0 and a 1 of
signal on the power lines that was beyond the ability of 0.024414 millivolts. This is arrived at by reducing the bits
to 11 to account for the plus or minus sign, then solving to pick up the smallest carbides and beta crystals possible.
the equation This is not to suggest that a single crystal event can be
seen, but rather that small clusters of crystals will produce
sufficient energy to be detected.
Equation 1
Smoothing of raw data
Precision = 50 millivolts / (2^11). Carrying this out for all
three configurations we get: To get the most out of a thermal analysis, the raw data
needs smoothing. With each successive derivative, the
Table 1. Millivolt accuracy of convertors noise is magnified, and the derivatives become more er-
12 bit convertors +/- 0.024414 millivolts ratic. Previous attempts by other researchers have been
16 bit convertors +/- 0.001526 millivolts blocked by the inability to get past the noise in the deriva-
tives, Previous attempts by other researchers have been
24 bit convertors +/- 0.000006 millivolts blocked by the inability to get past the noise in the deriva-
tives 11.
At the temperature of 50 millivolts, those variations added
to the standard signal would represent temperature varia-
tions of:

Table 2. Temperature accuracy of convertors

Pre- mV F C ∆F ∆C
cision
∞ 50.000 2250.24 1232.35 0.000 0.000
12 Bit 50. 024 2251.46 1233.03 1.212 0.678
16 Bit 50. 001 2250.31 1232.39 0.076 0.042
24 Bit 50. 000 2250.24 1232.35 0.000 0.000

Precision is then improved by the bits in the A/D proces-

sor. How much of this precision is just empty magnifica-
tion due to inherent noise is yet to be determined. Fig 3. An A319 curve with magnesium showing the un-
smoothed cooling curve. The red temperature curve
shows three arrests, while the green cooling rate curve
Capturing raw data shows additional arrests that were invisible on the temper-
th
ature curve. The black 4 derivative curve, which will be-
Raw data was captured by a commercial 16 bit unit and come important later, is currently too noisy to be useable.
the millivolts readings were saved in computer format
with no smoothing or filtering. The commercial unit has In figure 3, an unsmoothed curve shows clearly that there
an impedance value of 10 mega-ohms. This is important is additional information in the curve beyond the two ma-
because the higher the resistance, the less energy is con- jor arrests. In all the curves presented here, unless stated
sumed in the measurement, and therefore the measuring otherwise, the scaling is 1 degree F per second for the rate
unit alters the signal less. It is similar to Heisenberg’s of cooling, and the 4th derivative scale is 1/100 the cool-
Uncertainty Principle9 that says the very act of measuring ing rate.
changes that which is being measured. Electricians use
the term impedance to describe the effect of resistance There are many smoothing techniques that were investi-
along with the effect of capacitance10. This impedance can gated and tried12. These included a running average, clip-
generally be thought of as resistance for this paper’s sim- ping, local linear regression, etc. In general, the more
ple uses. smoothing done, the less detail is preserved. Most
smoothing looks at a group of data points and tries to plot
With the 10 mega-ohm impedance and at a value of 50 the most meaningful line through that data. A large group
millivolts, the instrument is then very sensitive to very size smoothes more, but a smaller group size preserves
small amounts of energy signals and has a low chance of detail. After a lot of trial and error, it was found that using
affecting a weak signal. Solving the equation I= V/ R several different techniques on top of each other while
where V = 0.050 volts and R = 1E10 ohms the solution is keeping the group size small gave the best results. The
5E-12 or 0.05 millionths of an amp are actually consumed best group size for a good clear signal worked out to be
in measuring the voltage. The amount of current being about 1 to 1.5 seconds of data, or for our instrument, 10 to
supplied depends on the surface area of the thermal 15 data points. If the signal contained external noise due
couple joint and the overall resistance of the system. This to poor shielding, then the group size had to be increased
will be important in the thermal analysis, because the size to 25 or more and some information was lost.
and quantity of the crystals being detected will control
how much energy is generated, and should be optimized
Figures 4 through 9 show the green cooling rate curve in
both the smoothed state and in the unsmoothed state. This
test was used to verify the validity of the smoothing algo-
rithm and judge how well this would fit a human who
could draw a freehand curve through the noise. Too much
smoothing deemphasizes the arrests as can be seen in the
circled arrest in figure 5. The level of smoothing is better
but still shows several small changes in the cooling rate
due to oxides. At this point, information has clearly been
lost and the smoothing needs to be reduced. Figure 4
shows the same curve with less smoothing and a more
pronounced 3 arrests. The 4th derivative however is now
very noisy.

In the example of figure 5 during the exothermic liquidus Fig 5. By increasing smoothing of the same area with a
there is little variation in the unsmoothed curve. While in group size of 25 the top oxide arrest has weakened consi-
weaker arrests and the non-arrest segments there are large th
derably but the 4 derivative is now noise free and the
swings in the curve from moment to moment. It is like the computer can pick up the arrests.
events producing energy are few and far between in the
non-arrest areas. In the minor arrest at points 1, 3 and 4,
there is a slight reduction in variation. At the strong ar-
rests at 2 and 5, there is a more pronounced reduction in
variation. A suggested explanation might be that the crys-
tal growth is like popcorn, explosively crystallizing and
then abruptly halting until the next crystal cluster pops.
During major growth, the crystal growth is so rapid that
the cooling rate naturally smoothes out.

Fig 6. Liquidus region of figure 3, an A356 alloy, showing

both smoothed and unsmoothed cooling rate (green).

Fig 7. Liquidus growth region of figure 6 showing beta

crystal arrest with both unsmoothed and smoothed cooling
th
rate in green and a smoothed 4 derivative in black. The
Fig 4. Smoothed oxide and liquidus region of base iron in size of the black curve inflections during the beta arrest
high detail with raw data (blue) versus group size of 10 are greater than 3 times the background variation to the
smoothing for the cooling rate. Due to a lack of shielding, left giving a 3 sigma confidence level of the reality of the
th
a noisy 4 derivative is shown in black (more smoothing is beta crystal arrest.
needed).
Fig 8. Grain Boundary of figure 3 showing both smoothed
and unsmoothed cooling rate data. Both the copper and
magnesium silicide arrests are tracked, through some over
smoothing is visible on the bottom of the copper arrest.
th
The 4 derivative is clean and it’s major zero crossings
indicate the beginning and ending of significant arrests.

Fig 10. A thermal analysis curve (red) and the cooling rate
(green) of an A319 aluminum. Purple circles indicate exo-
thermic crystallizations of Chinese script, liquidus, eutec-
tic, and a copper grain boundary. The black circle indi-
cates shrinkage, not visible on the temperature curve.

Fig 9. Eutectic area of figure 3 showing both smoothed

th
and unsmoothed cooling rate data as well as the 4 deriv-
ative. The shrinkage arrest is now clearly visible due to the
th th
4 derivative and based on the size of the 4 derivative,
the shrinkage arrest can be deemed statistically signifi-
cant.

The Liquidus ceiling points delineate the part of the curve

within 0.1 degrees of the liquidus growth temperature.
The longer this period is (4 seconds in this case) the more
Fig 11. A final hyper-eutectic ductile iron curve showing a
likely that the growth temperature is the true liquidus graphitic arrest and a standard deviation of 0.00193 during
temperature. the nodule formation period of the eutectic. This sample
was evaluated at 95% nodularity in the lab.
To recover data from noisy curves, it was found necessary
to configure the program to allow the smoothing to be As can be seen in figure 10, three arrests are visible in the
varied by derivative. Each successive derivative was cal- temperature curve, but an additional arrest and an endo-
culated from the smoothed previous curve. The exact thermic shrinkage arrest can be seen in the cooling rate
smoothing algorithm remains a trade secret, but others curve. Thus the cooling rate curve has more “magnifica-
should be able to produce an approximation of these re- tion” than the temperature rate curve alone.
sults.

Dissecting the thermal analysis curve

Most of the smaller reactions in thermal analysis are too
weak to raise or lower the temperature, but they can and
do change the rate at which the temperature is changing.
For that reason, these effects are more noticeable on a rate
of temperature change graph sometimes referred to as a
derivative graph. In a cooling curve, an exothermic reac-
tion such as crystal production and growth will slow the
temperature loss, while an endothermic reaction such as
shrinkage will speed the temperature loss. While profes-
sors usually study derivative graphs, the inverse of a de-
rivative graphic or the cooling rate is easier for most
people to understand. In the cooling rate graph, zero is Fig 12. An almost invisible bump on the temperature curve
(red rectangle) shows up clearly on the cooling rate as a
down instead of up and crystallization causes dips in the
vapor phase arrest in a copper phosphorus alloy used to
cooling curve pushing the cooling rate toward zero. promote nucleation in hyper eutectic Aluminum. The tem-
perature of the vapor phase suggests 15.02% phosphorus
content based on the copper-phos phase diagram.
A different example of small arrests is in a difficult cop- area between the zero curve and cooling rate curve gives
per phosphorus alloy. The alloy has a liquidus that cannot the amount of energy generated by crystallization. This
be used to calculate phosphorus content, but a phosphorus energy can be further broken down into the energy of the
vapor phase arrest (where internal vapor liquefies) is liquidus arrest and the eutectic arrest, and in this case the
strong enough to produce a small inflection on the cooling copper arrest. The ratioing of these energies give an ap-
rate. proximation of the volume solid of these constituents.
This volume solid could be further refined if the specific
heats of each constituent was known, but for now this is
the best approximation possible.

Fig 13. A different view of figure 3 with the 4th derivative

added in. Purple circles indicate thermal events. A new
arrest appears on the far right. Red circles indicate normal
th
background noise. 4 derivative magnification is 1000x of
scale. The third purple circle from the left shows a
void/gas bubble/shrinkage arrest. Due to its location on
the curve, it is probability a shrink. Fig15. The SSCRL and SSCRS are shown here in an
A319 alloy. The Zero curve is shown in violet.
In figure 11, the graphitic liquidus arrest is invisible in the
temperature curve, but shows as a slight bump on the A new term SSCR (steady state cooling rate) was devel-
cooling rate curve, and has 4th derivative fluctuations that oped to describe the flat areas of the cooling rate curve in
are 6 times the fluctuations during the quiet part of the certain areas. This is generally a sloping flat spot in the
curve. cooling rate where no major events are occurring. The
first one, the liquidus SSCR (SSCRL) is commonly found
in Aluminum but not in iron. It is believed that the cool-
ing effect of the cup sand cup used in iron prevents the
stabilization of the preliquidus area of the cooling

If the SSCRL is plotted backwards, it, defines an equation

describing the line seen in figure 16 before the liquidus
start.
Fig 14. The zero curve is displayed in violet over the cool-
ing rate of the curve in Fig. 13. The straight red lines indi-
cate energy area boundaries that are labeled.

By adding in the 4th derivative a 5th thermal event

(shrinkage) arrest shows up. The 4th derivative is a
powerful method to find the start and stop of small arrests
that are invisible on the temperature curve, but show up as
inflections on the cooling rate curve. The point at which
the 4th derivative passes through zero can be used as the
start and stop of small arrests. The point at which the Fig 16. The SSCRL line shown here is slightly offset
derivative passes downward through zero (negative) downward so as not to cover endothermic reactions in pre-
marks the beginning and ending of an endothermic arrest, liquidus region of A319 curve.
as can be seen in figure 14 for the shrinkage arrest. An
upward pass through zero marks the strongest point of This equation has some usefulness in calculating the
that arrest. For crystalization arrests, which are energy of pre-liquidus arrests such as gas and Chinese
exothermic, the zero passes are reversed. A zero pass script, and the “knee of the curve” leading into the liqui-
going positive marks the beginning and ending of the dus arrest as was pointed out by Dr Backerud 14.
liquidus and the beginning and ending of the liquidus
growth region in figure 14.

The final thermal analysis curve to describe is the zero

curve. This curve describes the cooling rate as if there
were no thermal events13. Integrating or measuring the
 The question arises: how can it be determine if an arrest is
real, or just a factor of the noise of solidification? With
larger arrests, it is clear that the data is real. There are 3 to
30 seconds of similar data points proving a strong arrest.
A more difficult challenge is to look at small arrests of
short duration and see what confidence level can be de-
termined for these arrests. For this analysis, the 4 th deriv-
ative proves again useful.

Fig 17. This graphic describes a complicated way of cal-

culating the true start of liquidus as explained by Dr.
Backerud. A tangent is taken from the liquidus start point Fig. 19. The rate of cooling and 4th derivative shown dur-
as defined by the last 4th derivative going negative and a ing eutectic solidification in ductile iron using 1/10 degree
th
second tangent is taken from the maximum of the last 4th scale for the 4 derivative.
rd
derivative before liquidus (same as the zero pass of the 3
derivative), and the intersection of those two lines is cal- As can be seen in figure 19, during the eutectic solidifica-
culated as the true start of liquidus. tion, the 4th derivative is effectively zero during the arrest.
If we can calculate a standard deviation on the 4th deriva-
A more complete explanation of this point is found in Dr. tive curve during this quiet period, it will be possible to
Backerud’s work. For this research project, the liquidus gage how far from the norm an arrest might be.
start point was used to begin the zero curve.

Standard Deviation equation 2

Fig 18. In this final ductile iron example, the Solidus is en-
dothermic because of stress or micro-shrinkage in the
grain boundary. The SSCRS terminates the zero curve
(violet)

The zero curve is a concept not fully defined as yet,

though Dr Kierkus15 of the University of Windsor has Fig. 20. A copper arrest, a magnesium silicide arrest, and
a small unknown arrest in A319 aluminum compared to
added much to its understanding. In aluminum the begin-
the quiet area of the curve (on the left).
ning of the zero curve should be defined as either the start
of liquidus arrest, or the true start of liquidus arrest if that
is found. The end of the zero curve should be at the start
of the Steady State Cooling Rate of a Solid (SSCRS). It is
at or after the solidus, but does not necessarily pass
through the solidus. Especially in ductile irons, there is a
sharp endothermic reaction at the solidus point. It is the Fig. 21 An example of gas formation in an A319 aluminum
author’s view that this is due to grain boundary stress. In a alloy. the square boxes before and after the gas/shrink
kind of reverse psychology, irons in which shrinkage ac- combination show the typical background noise. The ac-
tually occurs, have a low degree of grain boundary stress tual gas and shrink arrests are several times the standard
because the stress was released by the shrinkage. So high- deviation (std) of the background noise.
er stressed grain boundaries may indicate iron that did not
shrink. Figure 15 shows a simple version for aluminum,
and Figure 18 shows the difficulties in finding and defin-
ing the zero curve in iron where the start of liquidus is
often lost due to the rapid cooling rate in a sand cup, and
iron’s higher thermal loss due to radiation.

Measuring Normalcy
Fig. 22 An example of beta crystal formation in A319 Alu-
minum. The red box to the left shows the normal amount
of noise in the system during a quiet time. The sudden
 th
movement of the 4 derivative is 6 times the quite time Of course the other interesting fallout from this experi-
std. ment was the correlation between nodularity and the stan-
dard deviation of the eutectic area of the 4th derivative. In
In figure 22 there is a small aluminum arrest known as a the three final ductile iron examples (Figures 11, 23, 24),
beta crystal. This arrest again shows a 4th derivative that is the following results were obtained:
clearly greater than 3 times the normal noise in the 4th
derivative. While it is most likely that this arrest was gen-
th
erated by many of these crystals forming at the same Table 3. Standard Deviation of 4 derivative
time/temperature, the peak of the 4th derivative shows that during eutectic as a function of nodularity
the confidence level for the arrest being real is statistically Nodularity Std Dev
in excess of 99.7%. 95+% 0.0195
90% 0.0257
Ductile iron has a slightly different interpretation of the 80% 0.0490
eutectic normalcy. During this time, both spheroidal gra-
phite and vermicular graphite can be growing. Since they The graphite arrest in figure 25 has a value of about 6
grow by different mechanisms, diffusion for ductile and standard deviations greater than the normalcy value of the
110 edge growth for grey(some differences of opinion on quiet areas of the curve. The carbide arrest shown on the
this subject exist), their energy signals are different. This right of figure 24 is clearly greater than 2 standard devia-
leads to a noisier signature when vermicular is present tions from this rather rough eutectic, so the confidence
and a quieter signature when only the diffusion growth of level is greater than 95% that it is a true arrest. From basic
nodular graphite is present. foundry experience, such low nodularity irons often pro-
duce carbides.

TURNING THERMAL EVENTS INTO INFORMA-

TION

With a test for small arrests, the next step would be to

identify, define and measure these events for the benefit
of the foundry. Identifying is deciding what the cause of
the arrest is. If it is exothermic, it could be a type of crys-
tal, or a phase change. If it is endothermic, it could be gas
Fig 23. A final eutectic ductile iron curve showing good
or a shrinkage type of void. The type of metal, the tem-
nodularity (90%) and a standard deviation of 0.0257. perature of the event/arrest, the impurities, and previous
micros will help decide what each event is.

The first level of information is often just that the thermal

event is present. A graphitic liquidus in final ductile iron
indicates a hyper-eutectic arrest and a bimodal graphite
distribution. Likewise an austenitic liquidus in such iron
indicates a hypo-eutectic iron; while the presence of both
graphitic and austenitic liquiduses in a single iron indicate
a hyper-eutectic iron with a carbon flotation situation
Fig 24. This is a final hypo-eutectic ductile iron curve with (carbon segregation).
only minimum acceptable nodularity (80%), a standard
deviation of 0.0490, and a small carbide arrest. Other arrests that can indicate trouble with a heat include
shrinkage events, gas events, oxide arrests, which, if un-
treated, will reduce inoculation and magnesium recove-
ries, unwanted carbide arrests in final gray and ductile
irons, and beta crystals in 300 series aluminums.

Defining the event is finding its beginning and end. Mea-

Fig 25. This is an example of Hyper-hypo-eutectic ductile suring the event is noting both the temperature of the
iron having both a graphite liquidus and an austenite liqui- event, and the amount of energy produced or consumed.
dus arrest common in Hyper-eutectic irons above 4.6 If the amount of energy of an event and the total amount
16
C.E. . The appearance is slightly different than figures 23 of energy of solidification is known, then it is possible to
and 24 due to the change in the scale needed to include calculate an estimate of the volume solid of the event.
the graphitic liquidus.
But beyond just detecting the presence of microstructures
in a metal, these thermal analysis techniques hold open
the prospect of actually quantifying these microstructures.  Determine the total energy between the zero
The technique is to calculate the area of the arrest or void curve and the cooling rate by integration.
and ratio it against the overall energy of solidification.  Determine the typical standard deviation of a
sample in the area where no thermal events are
In trying to determine the start and stop of a minor arrest occurring, or define a typical standard deviation
such as in figure 22, the 4th derivative again has some for a process based on history.
welcome characteristics. The first upward pass of that  Find a 4th derivative maximum (crystallization)
derivative seems to be a good approximation of the start or minimum (void formation) that exceeds a user
of the beta crystal arrest, the first downward pass of the confidence level (95% or 99%) in the areas of
derivative seems to be a good approximation of the the curve where the arrest is typical.
strongest point of the arrest, though the second and third  Go back one 4th derivative crossover point and
derivatives may be better measures of the strongest point mark that point as the start of the arrest/void.
of the arrest or void. Still, the second upward pass again  Go forward one 4th derivative crossover from the
seems to be a good place to mark the end of the arrest. maximum/minimum point found and mark that
With these points it is possible to segregate the energy of point as the ending point of the arrest.
that individual arrest from the total energy of solidifica-  Take the values of the cooling rate at those two
tion, and then to ratio the two energies. points, calculate the equation between those
points, and integrate between the cooling rate
Likewise, figure 14 shows a void forming in the sample and the line equation between the two points.
that can be bounded by the same 4th derivative crossing
 Ratio the result against the overall energy of the
points. The void is endothermic but its negative energy of total solidification energy.
formation can again be ratioed to give a benchmark of the
shrinkage propensity of the metal.
Larger microstructures, such as those formed by the liqui-
dus, eutectic, and solidus arrests, are more easily defined
by the cooling rate, and the 2nd derivatives. The solidus
point is the easiest point to find, as the second derivative
passes through zero

Fig. 26 This is an example of a copper arrest in A319 alu-

minum showing the cooling rate in green and the 4th de-
rivative in black. In this example, the Mg-silicide arrest
interrupts the copper arrest. A line over the top of the rate
of cooling encloses the energy of the copper grain boun- Fig 27. This is an example of tellurium treated base iron
th
dary precipitation. Positive zero passes of the 4 deriva- showing an austenitic liquidus, the transition between li-
tive mark the beginning and ending points of integration. quidus and eutectic, and the eutectic arrest. The cooling
Negative zero passes mark the points of greatest strength. nd
rate is shown in green, the blue curve is the 2 derivative
and its zero crossover points indicate the liquidus and
th
In the example of figure 26, the magnesium silicide arrest eutectic arrests. The 4 derivative is shown in black and
interrupts the copper arrest and confuses the energy area. indicates the preliquidus oxide arrests.
But if the copper arrest end is extrapolated to the magne-
sium silicide end then the contributions of each arrest can In the example from figure 27, the 2nd derivative crossing
still be approximated. zero marks the points of maximums and minimums in the
cooling rate for major arrest points and the boundary be-
In the case of a shrinkage arrest, the cooling rate moves tween liquidus and eutectic.
gradually into the shrinkage preventing the sharp changes
in the 4th derivative seen before. This makes that type of
energy signature more difficult to detect. But the combi-
nation of the 3rd and the 4th derivatives can pick up the
inflection and give starting and stopping points for inte-
gration.

These techniques can give computer software the rules

necessary to automate the analysis of small microstruc-
tures.
Fig 28. An example of liquidus undercooling in an alumi-
num alloy and the growth temperature detection. The Li-
quidus ceiling points are 0.1 degrees C lower than the
ceiling temperature indicating how flat the ceiling is and Fig 30. An example of an A319 modified with magnesium
that it took 3 seconds to go through the growth temp. A and with a larger than normal amount of copper. The
nd
quick ceiling indicates that the cooling rate may suppress green curve is the cooling rate, the blue curve is the 2
the true liquidus. derivative and determines the arrest points at its zero
th
crossovers, and the black curve is the 4 derivative and
Sometimes the liquidus is strong enough due to under- determines the start and stop of the arrests.
cooling to pass through zero. In figure 24, the maximum
liquidus temperature is given when the cooling rate curve Addressing the previous question of what degree of preci-
(green) crosses zero in a positive direction. Dr. Backerud sion is best; it would appear that 16 bit data is accurate
has named this point the “liquidus growth temperature”. enough to pick up the swings in data while 12 bit data
This point may or may not be the true liquidus tempera- would lose the information of small arrests. The addi-
ture because rapid cooling may prevent the curve from tional cost of a 24 bit analog to digital converter is not yet
obtaining the maximum temperature. The two points to justified.
either side of the growth temperature in figure 24 are
listed as liquidus ceiling points. These points are 0.05 SUMMARY
degrees lower than the growth temperature, and the com-
puter has measured a time between them of 3.2 seconds. This paper has tried to show the following:
The longer this time is, the more surety there is that the  Thermal analysis, by the use of adequate shiel-
growth temperature has reached the true liquidus. ding, and innovative smoothing routines can re-
liability produce derivatives down to the 4 th
level.
 That through these derivatives very small events
can be seen, identified and measured.
 That through the 4th derivative, the beginning
and ending of these small events can be pin-
pointed so that the energy of that event can be re-
liability measured through integration.
 That this technology has application in iron,
aluminum, copper and magnesium alloys and
Fig 29. This shows a eutectic growth temperature in an
possibly many other alloys.
aluminum alloy. The time at eutectic growth temperature is  That arrests in iron such as oxides, gas, shrin-
27 seconds, indicating that this is most likely a true eutec- kage, carbides and others can be identified, and
tic temperature. that differences in the eutectic background signal
seem related to graphite.
Likewise the eutectic arrest can be either at or above zero,  That arrests in aluminum such as Chinese script,
such as in figures 18 and 19, or can undercool and reheat, liquidus recalesence, eutectic recalesence, beta
as in figures 21, and 23. In the first case, the upward cros- crystals, grain boundary precipitates, gas and
sover of the 2nd derivative indicates the strongest point of shrinkage can be identified and measured.
the eutectic, while in the second case, the upward cros-  That technologies such as the Zero Curve can be
sover of the rate of cooling indicates the eutectic growth applied to compare arrest energies to total ener-
temperature, and possible true eutectic. gies for an approximation of volume solid of dif-
ferent phases.
In Figure 30, the solidus is defined when the cooling rate
reaches a maximum, and thus the 2nd derivative (blue With the development of these technologies, the foundry
curve) passes through zero in a negative direction. A pre- industry might be able to move into a new form of quality
vious point close to the 494 second mark has the 2nd de- control that promises faster analysis with less operator
rivative curve touching or slightly passing through zero. sophistication.
The computer software may need some fine tuning to pick
out the correct point. CONCLUSION

Real time testing for microscopic structures by thermal

analysis is now becoming possible. Some care must be
taken to provide a strong, clear signal and to shield that
signal to help differentiate between noise and real events.
But given these enhancements, and accepting that the
sampling cup is not the casting, benchmarks can be meas-
ured to make sure the metal quality is correct by a semi-
skilled worker pouring a sample and then checking back
in a few minutes for the results.
 grain size indicator in the eutectic arrest of iron would
This also opens up the mode of solidification to analysis. relate to nodule count, this could be very interesting.
More than just chemistry, the solidification mode takes
into account the inoculation and cooling rates of the metal ACKNOWLEDGEMENTS
and therefore more of the actual physical characteristics
of the final castings. This paper would not be possible without the help of
many foundries as well as the early guidance of John
Iron samples generally take 3 to 4 minutes to finish cool- Weiss, when he was the technical director of Leeds and
ing to the solidus point, and aluminum samples generally Northrup Metallurgical Division, professor Richard Heine
take 5 to 7 minutes to reach solidus. Other alloys such as of the University of Wisconsin - Madison, William Shaw
copper and magnesium have also been successfully tested. of ICRI, Lyle Jenkins of the Ductile Iron Society, Avery
Pure and almost pure magnesium alloys tended to burn up Kearny, consultant, Dr. Carl Loper of the University of
the equipment, but sufficiently alloyed magnesium proved Wisconsin - Madison, Dr. Doru Stefanescu of the Univer-
to be practical. sity of Alabama – Tuscaloosa, now with Ohio State Uni-
versity, Dr. J.H. Sokolowski, and Dr. W.T. Kierkus of the
The test is relatively quick, requires a lower skill level, University of Windsor, Ontario, Dr. John Berry of the
takes very little time to pour a sample, and can have a University of Mississippi, Cees van de Velde, and many
standardize interpretation due to the computer. The diffi- others who have given encouragement to solve these
culty is that the sampling needs to be consistent and cor- problems and bring this technology to the current state.
rectly done. But the learning curve for that should not be
more than a day or two. For operations that have full time
micro technicians, the cost savings could be large. For
other foundries, the timely indication of small problems
and the standardization of the test should improve overall
metal quality and reduce scrap.

In iron it is presently it is possible to detect oxides, shrin-

kage, and carbides in a thermal analysis sample. It is poss-
ible to get a measure of the eutectic activity to compare
with graphite growth in both gray and ductile irons. In
preliminary tests, higher activity values correlate with
more undesired forms of graphite.

In aluminum it is presently possible to detect Chinese

script, liquidus recalesence to 0.05 degrees C (a measure
of inoculation), the presence of undesired beta crystals, an
accurate eutectic recalesence (a measure of the freshness
of the modification), the eutectic undercooling (a measure
of the overall modification), and the presence of gas and
shrinkage. With some refinement, it should be possible to
integrate the areas of the grain boundary constituents to
come up with the relative percents of precipitated copper
and magnesium silicide as well as integrate the shrinkage
areas to give a measure of the total shrinkage.

With a new method of measurement, the next step will be

to try this tool on the more complicated problems of mi-
crostructure as well as correlation with physical properties
in those irons. Improvements to the program will concen-
trate on quantifying the arrests by measuring the areas of
each arrest and correlating those areas with lab results.

As a final thought, the examples used here are by no

means all that can be found through thermal analysis.
There seem to be pearlite indicators as well as chill indi-
cators in the analysis of final iron, and there seem to be
grain size indicators in both aluminum and iron. From
simple experiments, the inoculation level of A200 series
(using CuP) is indicated in the liquidus grain size, and the
REFERENCES
1
J. Gruzleski “Microstructure development during metal-
castings” American Foundrymen’s Society 2000
2
“Proceedings of the conference on thermal analysis of
molten aluminum”, Cast Metals Institute, 1985.
3
“Ultrasonic velocity testing of nodularity in coupons”,
Ductile Iron Society Research project 42, Nov 2008.
4
http://en.wikipedia.org/wiki/Magnification
5
http://en.wikipedia.org/wiki/Antenna_(radio)
6
http://en.wikipedia.org/wiki/Ground_(electricity)
7
The Temperature Handbook Vol. 29, Omega Devices,
pg Z-19
8
The Temperature Handbook Vol. 29, Omega Devices,
pg Z-13
9
http://en.wikipedia.org/wiki/Heisenberg_uncertainty-
_principle#Heisenberg.27s_microscope
10
http://en.wikipedia.org/wiki/Impedence
11
Private conversation with Dr. J.H. Sokolowski .
12
Speyer, Robert F. “Thermal analysis of materials”,
Marcel Dekker, Inc 1994 pg. 95-97
13
W.T. Kierkus, J.H. Sokolowski “Recent Advances in
CCA: A New Method of Determining Baseline Equa-
tion”, AFS Transactions 1999, pg 161-167
14
L. Backerud, G. Chai and J. Tamminen, “Solidification
Characteristics of Aluminum Alloys,” AFS-
Skanaluminium, vol 2 (1990).
15
W.T. Kierkus, J.H. Sokolowski “Recent Advances in
CCA: A New Method of Determining Baseline Equa-
tion”, AFS Transactions 1999, pg 161-167
16
Cees van de Velde “The Hyper-eutectic Region of the
Iron-Carbon Diagram” http://members.lycos.nl/cvdv/
2004