The document summarizes an experiment conducted by Harrison Steel Castings Company to identify variables that affect surface oxide inclusions ("dirt") in steel castings.
The experiment collected data from 35 heats and identified 7 important variables related to dirt levels, including head height, pouring temperature, oxidizable element levels, and slag conditioner usage. Subsequent phases aimed to control these variables, introduce a new gating system, and measure the effects on dirt levels. Statistical analysis identified differences between "clean" and "dirty" heats for each variable. The experiment aims to better understand melting and pouring practices to further reduce dirt in castings.
1. Clean Steel Project— INTRODUCTION
The Harrison Steel Castings Company manufactures parts for off-
Identification of Variables road mining and earth moving equipment. The shop pours primarily
into green sand molds from bottom pour ladles. Melting furnaces
That Affect Cope Oxide consist of two 20-ton and one 8-1/2-ton acid-lined, electric arc
Inclusions in Steel Castings furnaces.
A clean steel project was started, in an effort to identify the
variables that influence the variation in cope surface oxide inclusions
J.D. Carpenter from one heat to the next. It was hoped that, by collecting as many
R.G. Shepherd melting and pouring variables as possible, statistical analysis could
Harrison Steel Castings Co. be used to identify the important variables that would explain this
Attica, Indiana variation. The following was stated in a paper given at the 1995 Steel
Founders’ Technical and Operating Conference: “It is our opinion
that reoxidation products are associated with every pound of metal
poured; however, factors such as casting design, gating system and
ABSTRACT
metal flow determine the degree and existence of these defects.”1
This paper discusses an experiment at a midwest steel foundry From previous work associated with the clean steel project, “ap-
concerning the surface cleanliness of one particular casting. proximately 83% of the casting defects in carbon and low-alloy steel
The experiment was started in an effort to identify variables that samples collected in the United States consisted principally of
have an effect on casting surface cleanliness. The emphasis on reoxidation products.” 2 The main reason for this project was that
the specific variables that were collected for the study was “…there is still an unknown set of circumstances, which exists in the
placed on the acid steelmaking and pouring operations. process, that suppresses the formation of reoxidation products or
otherwise modifies the nonmetallics such that the degree of visual
The experiment began by collecting as many variables as material is much reduced.” 1
possible from heats containing the casting of interest. The data
from the 35 heats was analyzed to isolate the important vari-
ables. The next phase of the experiment involved setting the
EXPERIMENTAL
identified variables at their optimum level for 24 heats, in order
to verify their effect. Inspection Methods for Castings
A change was then made to the gating system to reduce the The inspection methods used for examining the castings for defects
velocity of the stream into the mold cavity, and create a more were visual and magnetic particle inspection. Harrison Steel uses
streamlined flow. More data was collected by observing heats both wet and dry magnetic particle inspection methods. The first 29
with the new gating system. The collected data was reanalyzed trials used dry magnetic particle inspection. The inspection process
to try to confirm the significance of the original variables. was switched to wet magnetic particle inspection during the 30th
trial. The difference between the two processes was that, with the wet
Both steelmaking and pouring variables were found to have
magnetic particle inspection, an average of two inches fewer cracks
an effect on the heat-to-heat variation in surface casting quality.
per casting were detected. No effect on the amount of dirt was
The large within-heat variation suggested that the pouring
detected with the change to wet magnetic particle inspection.
variables have the greatest influence on the surface quality of
the castings. The measurement system selected for the study was the current
process for marking and recording defects after the first magnetic
Current trials are underway to gain a better understanding of
particle inspection. The quality control technician records each of the
the variables associated with melting practice changes. An
marked weldable defects in the appropriate location on a sketch of the
oxygen probe will be used to quantify the effect of the changes on
part being inspected. A length is associated with each marked defect.
active oxygen in the system.
The length assigned to a defect is the longest dimension of an oval
that can be drawn around the defect.
The total number of each type of defect is recorded, as well as the
total inches of each type of defect. The categories of defects are
cracks, dirt (sand or oxidation products), shrinkage, misrun, gas and
other. In order to reduce the variation in the rating process, all of the
castings associated with the project were marked by the same
inspector and recorded by the same technician. The goal of the
project was to reduce the defect category labeled “dirt.”
In this report, the dirt number is the number of areas where dirt
was found on the casting’s surface; the dirt length is the total severity
of the inclusions. A high correlation between the dirt number and the
dirt length can be seen in Fig. 1. This conclusion was important
because a reduction in dirt will reduce both the number and the
severity of defects. Dirt length showed the most variation and was
Fig. 1. Dirt number vs. dirt length. chosen as the dependent variable for the data analysis.
AFS Transactions 97-31 237
2. Test Casting
The casting that was chosen for the study was an engine support for
a large off-road vehicle that Harrison has been producing since 1976.
This casting is termed #2260 in this paper. Figure 2 shows the shape
of the casting. Figure 3 shows the gating system on the casting. The
gating system had approximately a 1:1:1 sprue:runner:ingate ratio.
The changes made to the casting during the previous 19 years are
shown in Table 1. Recent efforts had lowered the average dirt length
on the casting from around 42 inches in 1993 to around 32 inches in
1994. The clean steel data collection started at the beginning of 1995.
Each time this casting was run, five castings were poured in the
heat. Each casting had a gross pour weight of 2260 pounds. These
five castings accounted for 28% of the weight poured from a 20-ton Fig. 2. Shape of #2260 casting.
heat. The castings could be poured anywhere from the start of the heat
to the end of the heat; however, they should be poured back-to-back,
in order to keep the measurements for individual castings as close
together as possible. The nozzle could be burned out, but no other
castings could be poured between the molds.
Since the heat-to-heat dirt variation rather than the within-heat
dirt variation was being studied, the defect lengths from all five
castings were averaged. This number represented a cleanliness level
for all of the #2260 castings poured in the heat.
When a heat containing the #2260 casting was scheduled for
pouring, an observer was always notified. Most of the data used in the
study was normally recorded for every heat poured. The observer
ensured the integrity of the recorded data, collected the data not
normally recorded and collected the samples taken during the heat.
The observer also ensured that special instructions for each trial were
followed. The day following a heat, all of the gathered data was
assembled into a packet and assigned a sequential trial number.
Variables
The project of data collection started with a list of variables to be
collected that was assembled by the clean steel project steering
committee. This list consisted primarily of melting and pouring
practice variables. A complete list of the variables collected is given
in the Appendix.
The actual data collection started as soon as the heat was melted
in the furnace and continued until the ladle was slagged. Molding and
sand variables were considered background noise in the analysis of
the data, except when scabs were found on the gating system. One
heat with scabs on the gate was eliminated from the study, in order
Fig. 3. Original gating system.
to limit the analysis to reoxidation products rather than including
gross sand defects.
sample were used by the melter to calculate the adjustments in the
Melting and Pouring Practice ferroalloy additives necessary for the heat, based on the carbon,
manganese and silicon levels. After the block additions, the tempera-
The furnace was initially charged with bags of graphite to increase ture was adjusted and the heat was tapped into the ladle. The final
the meltdown carbon level. A charge containing approximately 50% deoxidation of the heat was done by adding pellets of aluminum to
foundry returns and 50% purchased scrap was then dropped into the the tapping stream at the rate of 2 lb/ton. The chemical specification
furnace. When the initial charge was partially melted, the balance of for the steel poured for the trials is listed in Table 2.
the heat was added as a back charge, if needed. When the bath was
completely melted, a chemistry was taken from a sample dipped from When the heat was tapped, a chemistry sample and a temperature
the furnace. The metallurgical lab reported the level of carbon, were taken in the pit. The chemistry was quickly checked to deter-
manganese and silicon to the melter. Oxygen was injected to lower mine if the heat was acceptable to be poured. If the heat was to be
the carbon to the desired aim point. After oxygen injection, a sample treated with calcium wire, it was treated in the pit.
was taken and sent to the lab to verify that the carbon aim had been The castings for the project were poured with a bottom-pour ladle
reached. through a two-inch diameter clay nozzle. The flow was controlled
After verification that the heat was ready to block, another dip with a graphite-tipped stopper rod. All of the ladles were lined with
sample was taken just prior to blocking the heat. The results of this bloating fireclay brick.
238 AFS Transactions
3. Table 1. After the pattern was rerigged for the basins, it was decided that
History of Changes to #2260 Casting the first few heats would be poured without the basins. Without the
pouring basins, the average dirt length for the first nine heats was 2.2
in. with a standard deviation of 1.745 inches.
Phase IV was an attempt to measure the effect of calcium wire on
the cleanliness of the castings by removing the wire requirement.
This phase consisted of data collected on heats with the new rigging,
without the injection of calcium wire in the pit. The average dirt
length for these heats was 9.3 in. with a standard deviation of 3.514
inches.
Phase V will be continued data collection using the new gating
system and calcium wire injection. A more rigorous investigation of
melting practice will be done using an oxygen probe.
STATISTICAL ANALYSIS
TECHNIQUES
The majority of the statistical analyses consisted of comparisons
between different groups of data. The average and standard devia-
tions were calculated for each variable in the two groups. Two mean,
equal variance test statistics were run to determine if there was a
difference in each variable between the two groups. Simple linear
regression analysis (least squares) was also used for continuous
variables. Pearson product moment correlation analysis was used to
Phases of Data Collection determine how variables affected one another (Note: only correla-
tions that exceeded the 90% statistical confidence level are shown in
Phase I was observation with little or no interference with the making
the paper.) Since the heat-to-heat variation was being studied rather
and pouring of the heats. After 35 heats were observed, the data was
than within-heat variation, the variables that were unique for each
analyzed using student’s “T” tests and regression analysis. The main
casting were averaged for each heat.
variables that had a potential effect on dirt (cope surface
macroinclusions) were identified. The average dirt length for these
heats was 32.3 in. with a standard deviation of 8.306 inches. Phase I
Phase II involved setting the variables, identified as being impor- The analysis was initially started after the first 12 heats had been
tant in Phase I, at their optimum value. These 24 trials were attempts collected. The benefit from beginning the analysis this soon was that
at making clean steel. The goal of these trials was to see if the it identified non-normal distributions for some of the variables. All
identified variables did, indeed, have the proposed effect on the dirt of the statistical work done on the trials required that each variable
length detected at the first cleaning-room inspection. The average have a normal distribution. The cause of the discrepancies was
dirt length for these heats was 19.8 in. with a standard deviation of related to comparing heats from the 20-ton furnaces with heats from
7.542 inches. the 8-1/2-ton furnaces. (Charge makeup and ladles were consider-
ably different.) It was specified that this casting could only be poured
Phase III came about in an attempt to separate the effect of head
in heats from the larger furnaces, and data that had been collected
height from pouring temperature. Pouring at the start of the heat was
from the smaller furnace was eliminated.
equivalent to having a high head height and a high pouring tempera-
ture. In order to separate the effect of these two variables to see which All of the data was entered into a spreadsheet and the heats were
was more significant, one had to be controlled. Controlling tempera- then ranked by average dirt severity. This ranking was then used to
ture was not a viable option; therefore, it was decided to use a pouring form two separate groups, labeled clean heats and dirty heats. The
basin on all of the molds. In order to add a pouring basin, yet keep the averages and standard deviations were found for all of the indepen-
same pouring times to avoid mix-running the castings, the pattern dent variables in the study. A student’s T test was then performed on
had to be rerigged. every independent variable collected, to find out whether or not there
was a difference in that variable between the groups of clean and dirty
heats. (The logic being that, if the variable had the same value when
Table 2. the heats were clean as when the heats were dirty, that variable could
Chemical Specifications not have had a major influence on dirt.) The output given for a
student’s T test is a confidence level that the averages for each group
are different. Each variable was assigned a confidence level and the
variables were then ranked by this confidence level.
The percent chance that there was a difference in the value of the
variable between the clean and dirty heats did not imply that there
was a relationship between the variable and dirt. It only implied that
one might exist. All of the variables that did not meet the 90%
confidence level were eliminated from the study.
AFS Transactions 239
4. Thirty-four variables remained in the study, with a confidence Phase II
level greater than 90%. A group of people, including representatives
The variables identified as being important in Phase I were set at their
from Steel Founders’ Society, the University of Alabama at Birming-
optimum levels in Phase II to try to make cleaner heats. A meeting
ham, Tri-State University and Harrison Steel, reviewed the data after
was held with the melters and pouring supervisors to explain what
24 heats had been collected. It was decided that many of the variables
had been learned. The melting practice was modified in order to aim
remaining in the study were representing the same thing. For in-
for higher levels of oxidizable elements after decarburization. The
stance, the levels of all of the oxidizable elements in the bath just prior
molds were all poured at the extreme end of the heat in order to satisfy
to blocking the heat remained in the list. Each of these variables were
the requirements of lower head height and temperature. The number
examined individually to see which had the highest correlation with
of bags of slag conditioner was increased from one or two to three or
dirt. The one with the highest correlation with dirt was chosen to
four. The melters were instructed to tap the heat in less than 88
represent that family of variables, and the rest were discarded. (Two
seconds. The number of heats on the furnace was allowed to fluctuate
variables were kept if the R 2 values were extremely high, as in the
because cost and time considerations made this variable impractical
case of block Si and Mn). Engineering judgment was used to reduce
to control.
the list to the seven important variables shown in Table 3.
After data from 24 heats was collected, the final analysis was
Figure 4 shows the distribution of dirt lengths for the 35 observed
performed. The analysis of the data consisted of the following.
heats. All of the variables will be discussed in detail in Phase II of the
project. 1. A comparison for each variable between the 10 heats with the
lowest dirt lengths against the 10 heats with the highest dirt
lengths.
Table 3. 2. A comparison between the dirt length for the highest 10 heats
Seven Important Variables from Phase I
against the lowest 10 heats of each variable (i.e., the highest
10 head height heats vs. the lowest 10 head height heats).
These two analyses were used as the basis for excluding or
including variables. All of the variables that had a statistical confi-
dence of over 90% in both studies are listed in Table 4.
Most of the variation in cleanliness on the #2260 castings could
be explained using three variables. The first two variables were
related to where the castings were poured in the heat. Pouring
location, from the start of the heat to the end, sets the level of the head
height and the pouring temperature in a bottom-pour ladle. The third
variable was the level of residual silicon in the bath after decarbur-
Table 4. ization (referred to as block silicon). The block silicon was highly
23 Important Variables from Phase II correlated with almost all of the variables during the making of the
heat.
Fig. 4. Phase I dirt length distribution.
Table 5.
Effect of Pouring Temperature on Dirt Length
240 AFS Transactions
5. The importance of pouring temperature is shown in Table 5. High Table 9 also shows that the temperature after oxygen injection
pouring temperatures corresponded with dirty heats. Various theo- was highly related to the meltdown Mn and Si. As the meltdown Si
ries may explain this correlation. One theory was that the higher was increased, the temperature after oxygen increased. The oxida-
temperatures superheat the mold, causing the steel to freeze more tion of one pound of Si provides 12,887 Btu. Manganese does not
slowly, thus allowing the inclusions time to agglomerate and float provide much heat of oxidation (3131 Btu per lb5); however, it shows
toward the cope surface. Another theory is that the inclusions formed as being correlated, due to its high correlation with Si (when the Si
at the higher temperatures may be more detrimental to cope surface is high, Mn is high). The meltdown carbon did not correlate well with
cleanliness than inclusions formed at lower temperatures. Turkdogan, the temperature after oxygen. The oxidation of one pound of carbon
for instance, has shown that the compositions of the reoxidation generates 4375 Btu.5
products change as the formation temperatures change.3
Residual Cr had an effect on the level of residual silicon and Mn
The importance of head height in steel cleanliness is shown in after the injection process because of the inter-relationship between
Table 6. Previous work with the clean steel project used water meltdown Mn, Si and Cr. The level of block Cr for the grade studied
modeling to show that the air entrainment was proportional to the set the level of the final Cr, since there was no Cr addition. When final
head height, raised to the 2.5 power, times pouring time.4 Cr showed up in the list of 25 important variables, it related back to
heats that finished with a high block silicon.
With other items, such as inspection criteria and gating system
constant, the combination of head height and pouring temperature Table 10 shows the correlation between chemistries taken from
was the largest contributor to steel cleanliness for a casting. the furnace and the percent recoveries for the elements. The block Mn
and Si are highly correlated with the recoveries of these elements.
The second largest contributor to steel cleanliness was the biggest
The block chemistry and recovery for these elements were directly
surprise in the study. The importance of the levels of residual silicon
related to the temperature of the bath after oxygen injection. The only
and manganese, after the blow, came as a shock to the investigators.
Table 7 shows that higher levels of residual silicon, after the blow, are
Table 8.
associated with cleaner castings. Analysis of Variance Using 3 Important Variables
Analysis of Variance
Due to the large variation attributed to these three variables—
pouring temperature, head height and block silicon—an analysis of
variance was run on the 59 trials collected. Dirt length was the
dependent variable and head height, pouring temperature and block
silicon were the independent variables. All factors were found to be
significant. The results are shown in Table 8.
Oxidizable Elements and Recoveries
Table 9 shows the correlation between the chemistries taken from the
Table 9.
furnace. The meltdown chemistries were highly correlated with one Chemical Correlations
another, as expected. If the charges were built with higher alloy
materials, the levels of C, Mn, Si and Cr were elevated at meltdown.
The melt chemistry was also highly correlated to the block chemistry.
Higher meltdown chemistries had higher block chemistries. Cleaner
castings were obtained from heats with higher levels of oxidizable
elements after decarburization.
Table 6.
Effect of Head Height on Dirt Length
Table 10.
Recovery Correlations
Table 7.
Effect of Block Silicon on Dirt Length
AFS Transactions 241
6. recoveries that had an influence on the dirt length were the recoveries High Level of Oxidizable Elements
of Mn and Si. As the recovery of the elements increased, the dirt Prior to Blocking the Heat
length decreased. The authors believe that the correlation between a high block
chemistry and low amount of dirt must relate back to the oxygen
The recovery of carbon showed a correlation with the level of
availability in the bath after the blow. The recoveries were higher
carbon just prior to blocking the heat only. The recovery of carbon
when the block silicon was high. As the available oxygen was
was not correlated with the recoveries of Mn, Si or Al. Carbon
lowered (at higher block chemistries), the recovery of the elements
recovery was not associated with dirt severity.
increased and the dirt length decreased.
The recovery of Al was mildly correlated with the level of carbon
Much research has stated that oxygen content after the boil
in the bath. A separate analysis was made in order to examine what
follows a steady-state curve, based on the carbon content.6,7 From the
variables were affecting the recovery of Al, since no correlations to
data collected, one conclusion has to be drawn from the following
the other recoveries were found. Aluminum recovery was correlated
two choices: 1) the active oxygen in the bath was dependent on more
with the level of carbon in the bath, the tap temperature, the type of
than carbon content or 2) the oxygen available that affected the
calcium wire used (solid calcium vs. Cal-Sil) and the temperature
recoveries did not come from the steel bath. Work done by Fitterer
during the wire injection. The authors suspected that the length of the
states that the level of active oxygen in the bath, after the blow, is
tap, in seconds, would be correlated to the Al recovery, but no
dependent on more than the carbon content.8 According to Svoboda,
correlation was found. The shape of the stream during the tap would
“Silicon content after the carbon boil will vary from about 0.02 to
probably explain some of the variation in Al recovery and in dirt
0.10%, and Mn from 0.07 to 0.15%. Recommended aims at this point
length; however, a method to rate the stream consistently, from tap
are 0.08% Si and 0.15% Mn. Manganese content greater than 0.15%
to tap, was never found.
tends to inhibit gas removal because the Mn causes the boil to be less
vigorous.”9
Table 11. The levels of residual Si and Mn were not set by the level of the
Relationship Between Carbon and Recoveries carbon in the bath. They were directly related to the temperature of
the bath after the oxygen injection. Clean heats with high recoveries
were made at all levels of block carbon (0.085–0.195% C) as shown
in Table 11.
Oxidizable Elements and Block Silicon Levels
In order to further verify the importance of the block silicon, all of the
heats collected in Phases I and II were divided into two groups, based
on the block silicon level. The first group was all of the heats with
0.045 block silicon and higher (specified as a good melting practice).
The second group was all heats with less than 0.045 block silicon
(specified as a bad melting practice). Table 12 shows the values of
some of the related variables.
The primary difference between a good and a bad melting
practice was the temperature of the bath, during and after the oxygen
injection. The other differences are reflections of this variable.
Carbon removal is more favorable at the higher bath temperatures.5
The higher meltdown chemistries created higher bath temperatures,
Table 12. due to the heat released during the oxidation of Si and Mn. The higher
Good Melting Practice vs. Bad Melting Practice chemistries required more oxygen to remove these elements. As a
result of the higher temperatures, the slag became thicker and
required more lime and ore additions to get the viscosity back within
the specified limits.
The increased levels of oxidizable elements at the block was
believed to be related to lower levels of active oxygen in the system
available to oxidize the alloy additions. As a consequence of the
lower oxygen availability, the recoveries were higher. All of the
block alloy additions had to be lowered to keep the chemistry near the
aim points, due to these higher recoveries.
Two approaches were used to achieve a better melting practice
with higher temperatures during the oxygen injection. The first was
to raise the amount of oxidizable elements in the bath, such that the
excess heat from oxidation of these elements ensured that the bath
temperature was high enough. The second was to make sure the bath
temperature was high enough before oxygen was started. Both
practices have worked successfully; however, a combination of both
seemed to work the best to guarantee the proper temperature was
reached. A balance was found between being hot enough to have a
242 AFS Transactions
7. “good” melting practice and being too hot (which caused furnace Ladle Treatment Variables
lining damage and a thick, chunky slag). Currently, the melters are Early in the course of these trials, there was a change from steel-clad
obtaining the carbon, Mn and Si rather than just carbon before calcium wire to steel-clad Cal-Sil powder wire, for cost reasons. A
adjusting the temperature prior to oxygen injection. study was run in order to determine if there would be a detrimental
effect from the change. It was found that the Cal-Sil powder wire
Good melting practice was achieved with the following methods:
produced a much less violent reaction than did the solid Ca wire. As
1. Started with higher meltdown Si (increased by 0.11% Si).
a result of less steel exposed to the air during the injection process,
2. Started with higher meltdown Mn (increased by 0.15% Mn).
the Al recovery improved from 38.2% to 40.1% (90% confidence
3. Started oxygen with a higher bath temperature to promote
level). The Mn and Al losses per minute were also reduced with the
oxidation of carbon. (Starting temperatures were set based on
change in wire type (99.95% confidence level).
the meltdown chemistry.)
Results of a good melting practice were: The silicon recovery went from 87.6% to 94.7% (99% confidence
1. Finished with higher block Si (higher by 0.016% Si). level). Silicon is part of the addition with the Cal-Sil wire. Part of the
2. Finished with higher block Mn (higher by 0.038% Mn). Si addition was removed from the block and accounted for via the
3. Had higher recovery of Si (7.5% higher recovery). wire injection. By adding the last part of the Si addition to a fully
4. Had higher recovery of Mn (3.4% higher recovery). killed steel in the ladle, the recovery was much higher, as expected.
5. Had to add 26.2 lb less FeSi (also change in wire practice,
Pouring Practice Variables
shown later).
From earlier analyses, it was found that head height, pouring tem-
6. Had to add 46.5 lb less Si-Mn.
perature and block silicon account for a large percentage of the
7. Had to add 17.2 lb less Fe-Mn.
variation in casting quality. The effects of the other pouring variables
8. Had a higher hardenability due to Mn recovery.
are shown in Table 14. The flow rate was measured in pounds per
Discussion of Ability to second (gross weight poured taken from scales and divided by the
Measure Block Silicon Level number of seconds to pour mold). The stream rating was a visual
After reporting these findings, Harrison’s capability of measuring Si rating of the stream quality. A numbering system was employed that
levels in these ranges came under question. Two studies were rated the stream quality to a number from 1 to 5. Stream ratings of 1
conducted to determine the capability to measure silicon at these were considered perfect streams with no slant or flaring. Slanted,
levels. The spectrometer reports the value of the elements to multiple flared streams that were splashing out of the pouring cup onto the top
decimal places and then truncates them to five decimal places. of the mold were rated at 5.
The first study was a comparison between the value originally The conclusions from this data were that the most important
reported for block silicon with a new analysis run on the same variables were still the pouring temperature, head height and block
sample. Ten heats were retested and the new value found for Si was silicon. The highlighted conditions show that the best castings were
compared to the original value. The average difference in the Si poured with a good quality stream into the pouring cup and low flow
between the heats was 0.00136% Si. This accounts for about a 3% rates into the mold cavity.
difference.
Slag Thinner Variables
The second test was run on one heat with a low reported block The number of bags of slag thinner that was added to the ladle was
silicon and one heat with a high reported block silicon. Replicate tests found to have a significant effect on dirt length. Table 15 shows the
were run on each heat to see how much spread existed in the readings. effect of the slag thinner. This product was used by the foundry to
The average and standard deviations are shown in Table 13. The table keep the ladles cleaner. When the ladles were dumped at the end of
shows that the level of the residual block silicon after oxygen the heat, this addition allowed almost all of the slag to be removed.
injection was measurable to at least three decimal places and that a The specifications on the product are shown in Table 16. This
difference did indeed exist between heats. variable initially appeared due to differences in the addition practice
between the day and night shifts. Figure 5 shows that the slag thinner
Table 13. affects the dirt length regardless of where the castings are poured in
Measurement Capability of Block Silicon the heat. Heats with three and four bags of added thinner produced
cleaner castings than did heats with one or two bags.
Table 14.
Effect of Flowrate and Stream Quality Rating
AFS Transactions 243
8. Table 15. 2-Factor, 2-Level Designed Experiments
Slag Thinner’s Effect on Dirt Length Effects of Refractory Variables—There was an interest in what could
be learned from 2x2 matrices, with regard to how the refractory
variables interacted with other variables. The conclusion from this
analysis was that the importance of some variables was dependent on
the level of another variable. The effect of a variable was often
masked by the levels of other variables.
The effect of the number of heats on the furnace lining was
found to be significant in the original observation trials. As more
data was collected, this variable dropped out of the picture. The
number of heats on the furnace lining may have (70% confidence
Table 16. level) a very small effect on dirt length (less than 1 inch of dirt, on
Slag Thinner Specifications average). Clean steel can be made with a high number of heats on the
furnace lining.
The number of heats on the ladle refractories did not have an
effect on dirt unless other more important variables were at their
optimum level. The number of heats on the walls and bottom of the
ladle had a small influence on dirt in the cleaning room. At high head
heights and pouring temperatures, the effect of the ladle refractory
was masked. As castings became cleaner, the effect was more
pronounced. Table 17 confirms that the number of heats on the ladle
refractory was not a significant contributor to dirt unless the castings
were poured at the end of the heat.
Effects of Tapping Rate—No effect on the dirt due to the tapping
rate was found. This variable showed as being a potentially signifi-
cant variable in Phase I, but was found not to be a significant factor
during Phase II. The tapping rate has an effect on other variables, but
does not have an effect on the dirt found on the castings, as shown in
Table 17. Table 18.
2x2 Matrices on Refractory Variables
Distribution of Dirt Lengths
When the most influential variables were set at a more optimal level,
there was a wide distribution of dirt lengths. The average dirt length
was shifted downward, but the standard deviation was unchanged.
This wide distribution revealed that all of the variables that affect the
cleanliness were not being closely controlled. Figure 6 show the
distribution of dirt lengths from Phases I and II.
Phase III
Phase III came about from an effort to separate the effect of head
height from pouring temperature. Since bottom pour ladles were
used, castings poured at the end of the heat had a low head height and
were also the coolest, due to the longer times in the ladle. By putting
a pouring basin on all of the molds, the effect of head height in the
Table 18. ladle could be reduced. A better comparison could then be made
2x2 Matrices on Tap Rate between castings poured at the start of the heat (hotter) and castings
poured at the end of the heat (colder).
In order to see if this was feasible, one heat was poured with a
pouring basin on two of the five castings in the heat. The pouring time
doubled on the molds with the basins, and the castings were severely
misrun. In order to get a similar pouring time, the foundry engineers
had to open up the choke in the gate. The problem was that the choke
was the entire gate system (approximately a 1:1:1 gating ratio).
The casting was rerigged using a streamlined gate with a 1:3:3
gating ratio. The new gating system is shown in Fig. 7. Castings were
then poured with the new gate, without using pouring basins. Nine of
these heats were indexed and the average dirt length dropped to 2.2
inches, as shown in Fig. 8.
244 AFS Transactions
9. Fig. 5. 2x2 matrix on the effects of slag thinner.
Fig. 6. Phases I and II dirt length distributions.
Fig. 7. New gating system.
Phase IV
After data from nine heats with the new gate was collected and the
improvement was verified (2.2-in. average dirt length), the calcium
wire injection requirement was removed. This phase of the trial
existed for two reasons. The first was an academic interest in the
effect of calcium wire on the job. The second reason was an effort to
offset the additional cost to produce the casting after the rigging was
modified, which added 200 lb to the gating system. The effect of Ca
wire is shown in Fig. 9.
After data from four heats with the new gate without wire
treatment was collected, the requirement of calcium wire was rein-
stated. The average dirt length for the four nonwired heats was 9.3
inches. Figure 10 shows the distribution of dirt lengths for the four
phases of trials. One interesting note on this portion of the trial was
that the head height and pouring temperature have the same effects
found earlier. The two heats where the castings were poured at the
Fig. 8. Phases I, II and III dirt length distributions. start of the heat had more dirt, as shown in Table 19.
AFS Transactions 245
10. Phase V Table 19.
No Wire, Head Height and Pouring Temperature Effects
Data collection is continuing with the new gate system with the
calcium wire requirement, to see whether or not the variation in
cleanliness can be explained with the new gate using the same
melting and pouring variables. Oxygen probe analysis will also be
made on heats, in order to verify the suspected relationship between
residual block silicon and soluble oxygen in the bath.
CONCLUSIONS
1. It appears that the magnitude of the cleanliness was set by the
gating system (new gate vs. old gate). This was the most important
variable that determined how clean a casting was going to be, before
the first casting was poured.
2. A large variation in average casting quality was seen from heat to
heat, with a particular gating system. Some heats were cleaner than
others, in terms of casting quality. The heat-to-heat variation of
average casting quality was partially explained through both melting
and pouring variables.
3. The largest variation from heat to heat was correlated to varia-
tions in the head height and pouring temperature. Fig. 9. Effect of Ca wire.
4. The second largest variation from heat to heat was due to
differences in the melting practice that can be associated with the
residual block silicon level.
5. The large within-heat variation indicated that melting practice
could not account for all of the variation. Some of the variation must
be explained by the pouring variables. A similar conclusion was
drawn in a paper comparing acid vs. basic melting: A high variation
within a heat relative to the variation between heats suggests that
gating and pouring operations are the most important factors.10
6. The large within-heat variation, after controlling the head height,
pouring temperature and melting practice, was attributed to variables
associated with the filling of the mold cavity (flow rate and stream
quality).
Fig. 10. Phases I, II, III and IV dirt length distributions.
SUMMARY
Data collection and analysis of this type will continue to be an
REFERENCES
important part of daily operations at Harrison Steel. The knowledge
gained through these type of studies is invaluable to remaining a 1. J.E. Parr, “Clean Heat/Dirty Heat, The Practical Implications of Data
competitive supplier of steel castings. The data collected also serves Collection,” Transactions of the 49th SFSA T&O Conference (1995).
as a valuable base of information to find relationships between 2. J.A. Griffin and C.E. Bates, “Development of Casting Technology to
seemingly unrelated variables. We will continue to pose questions Allow Direct Use of Steel Castings in High Speed Machining Lines,”
that this data might answer. SFSA Research Report No. 100, May 1987.
3. E.T. Turkdogan, “Deoxidation of Steel—What Happens From Tap to
Further data collection will continue. If the melting and pouring Solidification,” Electric Furnace Proceedings (1966) pp 22-28.
practices have an effect on all gating systems, an emphasis will be 4. M. Blair, R. Monroe and C. Beckermann, “The Effect of Pour Time and
made on improving and more closely controlling both areas, in order Head Height on Air Entrainment,” Transactions of the 47th SFSA T&O
to get a better product. If the melting and pouring practices are only Conference (1993).
significant on certain gating arrangements, the gating practices will 5. BOF Steelmaking, Volume 2, Design Part II, Operation, Special Topics,
Process Technology Division, Iron & Steel Society of AIME (Chapter
have to be analyzed for their effects on cleanliness.
13).
Future trials will try to use an oxygen probe to determine if more 6. D.L. McBride, “The Physical Chemistry of Oxygen Steelmaking,”
consistent recoveries and better control of deoxidation can be accom- Journal of Metals, vol 12, No. 7, July 1960, p 531.
plished. The probe will also be used to try to correlate the amount of 7. D.A.J. Swinkels, S.R. Richards, C.J. Cripps Clark, C.W.P. Finn and A.T.
active oxygen in the bath with heat cleanliness. Hart, “Oxygen Probe Applications in Steelmaking,” Open Hearth Pro-
ceedings, vol 25 (1972), pp 80-93.
8. G.R. Fitterer, “Some Current Concepts of the Oxidation and Deoxida-
ACKNOWLEDGMENTS tion of Liquid Steels,” AIME Electric Furnace Proceedings, vol 35, 1977
pp 302-307.
The authors would like to thank Harrison Steel for allowing us the 9. J.M. Svoboda, “Melting and Deoxidation of Cast Steels” Steel Casting
opportunity to work on this project. Special thanks should also be Metallurgy, Steel Founders’ Society of America, 1984.
extended to B. Trimble, S. Hughes and T. Hanson for their efforts in 10. R. Monroe, M. Blair and J. Griffin, “Variations in Casting Quality
the data collection. Analysis” Transactions of the 49th SFSA T&O Conference (1995).
246 AFS Transactions
11. APPENDIX Warm Up Variables
Warm up Time . Time in minutes that it takes to warm up the heat to the
desired tap temperature after the block addition has been
made.
Variables Collected for the
Tap Temperature Tap temperature is taken just prior to rolling the furnace
Clean Steel Project over for tapping. The temperature is taken with an immer-
sion thermocouple.
Background Variables Pit Variables
Trial Number . . Sequential number to uniquely identify each trial Seconds to
Date . . . . . . . . . Date castings were poured Tap Heat . . . . Clock was started as soon as the first material came off the
Heat Number . . Heat identification end of the spout. The clock was stopped when the material
flowing from the spout was all slag and had tapered off
Heat Code . . . . . Coded identification for customer
considerably. The tap rate was calculated based on the
Serial Number . Uniquely identified each casting poured in a heat
steel weight tapped and the tap time.
Steel Grade . . . . Harrison Steel designation
Ladle Chemistry The ladle chemistry was dipped from the furnace with a
spoon. No additions were made to the test. The test was
Chemical Variables
dipped out as soon as the ladle was filled. The test was a
Chemical
“Go” or “No-Go” test for elemental ranges. Recorded
Analysis . . . . C, Mn, P, S, Si, Ni, Cr, Mo, Cu, B, Sn, Al, Ti, V, Zr, Dl
elements were C, Mn, P, S, Si, Al.
Melting Variables Pit Temperature Pit temperature was taken as soon as the ladle test had been
taken by an immersion thermocouple.
Furnace . . . . . . . Identification of melting furnace
Melter . . . . . . . . Melter’s name
Wire Feeding Variables
Heats on Furnace Number of heats on furnace refractory
Before Wire Cool The number of minutes the heat was cooled in order to get
Repairs After . . Whether or not the furnace lining was repaired after the
to the aim wire feeding temperature.
heat was tapped
After Wire Cool The number of minutes the heat was cooled in order to get
Repairs Before . Whether or not furnace lining had been repaired before
closer to the desired foundry temperature.
charging the heat
Type of Wire
Refining Variables Used . . . . . . . Wire type was either Cal-Sil (which was a steel clad wire
with solid Cal-Sil powder inside) or calcium (which was
O2 Volume . . . . Volume of oxygen injected into the bath, measured in
steel clad solid calcium).
standard cubic feet
Length of Wire
O2 Time . . . . . . Minutes of oxygen injection
Used . . . . . . . The length of wire added measured in feet. Used to
C Blown . . . . . . Total C removed (%) and % removed per minute of
calculate the amount of calcium added to the heat.
oxygen
Velocity of Wire The velocity of the wire into the ladle determined how
Mn Blown . . . . . Total Mn removed (%) and % removed per minute of
deep the wire gets before the steel cladding melted and the
oxygen
reactive metal was released.
Si Blown . . . . . . Total Si removed (%) and % removed per minute of
After Wire Temp The temperature was taken immediately after wire injec-
oxygen
tion had stopped.
Cr Blown . . . . . Total Cr removed (%) and % removed per minute of
Calcium per Ton Weight of calcium added to the heat by the heat weight in
oxygen
tons.
Block Chemistry Complete chemistry taken from dip sample just prior to
adding the block addition to the furnace
Ladle Variables
Furnace Addition Variables Ladle Number . . A serial number was used for each ladle as identification.
Dust . . . . . . . . . Weight of the recycled furnace dust added to the charge. Empty Ladle Wt. This weight was read off of the scales after the heat was
poured and the slag had been dumped from the ladle.
Lime . . . . . . . . . Added to condition slag, estimated weight of lime (in
pounds). Estimation is based on the average weight of a Heats on Walls . The number of heats that had been poured from the ladle
shovel full of lime. since it was rebricked.
Heats on Bottom The number of heats that were on the ladle refractory in the
Iron Ore . . . . . . Added to assist the Boil-Weight (in pounds) added, in
bottom. Rammed material. The ladle nozzle was changed
order to assist in getting a hard carbon boil in the furnace
after every heat.
during refining. Weight is based on the weight of an
Nozzle Size . . . . Diameter of the nozzle used in inches (all heats had a 2
average shovel full of iron ore pellets.
inch nozzle).
Pig Iron . . . . . . . Added to raise carbon level when 5 or more points of Ladle Lining . . . All of the heats were poured with ladles that had been lined
carbon are needed, pig iron is used. If less than 5 points of with bloating fireclay brick.
carbon are required, graphite is added to the stream during Nozzle Material All heats were poured with clay nozzles.
tapping of the heat.
Bags of Slag
Si-Mn . . . . . . . . Weight of silico-manganese added to the heat during the
Thinner . . . . . Number of bags of slag thinner used to assist in cleaning
block in the furnace.
the ladle of slag after the heat has poured. If one bag was
Fe-Si . . . . . . . . . Weight of ferrosilicon added during the blocking of the added, it was thrown on top of the slag after the heat was
heat in the furnace. tapped. If two bags were added, one was thrown in at 1/2
Med C Mn . . . . Weight of medium carbon manganese added during the ladle and the other after the slag was all tapped.
blocking of the heat in the furnace.
Extra Med C Mn When a long warm up time is needed, an extra amount of Foundry Variables
medium carbon manganese was added to replace the Foundry Cooled Number of minutes that heat was cooled in order to get to
manganese that was lost due to oxidation. the aim pouring temperature.
Graphite . . . . . . Weight of graphite added to the tap stream to bring the Foundry Time . . Minutes from tap until the ladle arrived in the foundry.
carbon level to the desired endpoint. Recorded to keep track of how long it had been since the
Aluminum . . . . . Weight of aluminum added to the tap stream to finish heat was tapped for temperature losses.
killing the heat and leave enough free aluminum to meet Foundry
the desired aim point. Temperature . Temperature in degrees Fahrenheit.
AFS Transactions 247
12. General Foundry Variables Height of Steel . Based on the geometry of the ladles, the height of the steel
Heat Director . . Name of the heat director in charge of the pouring opera- in the ladle was calculated before every pour. The height
tion. was used for velocity. Since there was only a small change
Ladle man . . . . . Name of person operating the stopper bar on the ladle. in height for each mold, the height at the start of the pour
Tap to Pour . . . . The time was recorded from the time the tap started to the was used each time.
time the ladle was opened for each box. The time was Head Height . . . The head height, in this case, was the height from the
recorded to the nearest tenth minute. bottom of the pouring cup (in the runner) to the steel level
in the ladle.
Pattern Specific Pouring Variables Volume of Steel Calculated volume of steel in ladle before each pour. The
Mold volume was in cubic inches and converted into gallons.
Identification The data was collected for all of the 2260 molds poured in Volume Poured . Volume poured for each box.
the heat. Weight of Steel . Weight must be calculated because the scale weight does
Pouring Order . . During the pouring of the heat, each event got numbered. not account for the weight of empty ladle. The weight of
The first event was always 1 and was the pouring of the the slag was unknown until the end of the heat.
initial ingot. The next event was the first mold poured. Slag Weight . . . Weight of the slag in the ladle.
Burnouts and tests were also recorded as an event. All
events got a time and weight recorded. Mid-Pour Variables
Weight Poured . Number of seconds to pour a mold. This time was recorded Mid-Pour Time . Minutes from the tao until tests were poured.
by the heat director for each mold. Mid-Pour
Flow Rate into Temperature . Temperature in ladle at the time the tests were poured.
Mold . . . . . . . This rate was calculated by dividing the gross pour weight Final Chemistry The final chemistry was the official analysis for the heat.
by the seconds taken to pour the mold.
Velocity of Oxidation & Reduction Variables
Stream . . . . . . This was calculated based on the square root of 2gh, where Manganese Fade The amount of manganese lost due to oxidation. The loss
g was 32.2 ft/sec2 and h was the height the steel falls before rate was calculated on a per minute basis to account for
reaching the cup in the runner system. different testing times for each heat.
Temperature at Aluminum Fade The amount of aluminum lost due to oxidation. The loss
Pour . . . . . . . This temperature was found by extrapolation from known rate was calculated on a per minute basis to account for
temperature and times. different testing times for each heat.
Burnout of Silicon Pickup . Silicon pickup from the ladle test to the final test. This loss
Nozzle . . . . . . If the nozzle was lanced with oxygen prior to the mold is also divided by time to account for different testing
being poured, the mold was labeled as a burnout mold. If times.
the nozzle was not lanced prior to the pour, the mold was
marked as none. Miscellaneous Variables
Visual Stream Ladle Slag . . . . . A visual examination on the color of each slag sample was
Rating . . . . . . Visual stream ratings were taken every 5 seconds during performed and the appearance of the ladle slag fracture
the pour. Each rating was a number from 1 through 5. A 1 was recorded.
was a perfect stream and a 5 was the worst stream. All
Furnace Slag . . . A visual examination on the color of each slag sample was
readings taken on each box were averaged to have one
performed and the appearance of the furnace slag was
number to represent the stream quality. A visual rating
recorded.
system was employed to characterize the integrity of the
stream during the pouring of each mold. A number from Points Added . . The weight percent of C, Mn, Si, Al and Ca were calcu-
1 to 5 was assigned to rate the stream. A 1 represented the lated based on the alloy additions. The points of each
best stream possible. A 5 represented the worst stream element were then divided by the heat weight to get the
possible, as far as flaring. The stream was rated every 5 points of each element per ton of steel.
seconds during the pour. At the end of the pour, all stream
Response Variables (Index)
ratings for one box were averaged.
Crack Number . Number of areas on the casting that require repair due to
Variables Measured or Used in Calculations cracks.
Gap (Nozzle Crack Length . . Length of circled areas on the casting that require repair
to Can) . . . . . This measurement was necessary to calculate the total fall due to cracks.
of the stream for velocity calculations. After a few trials, Dirt Number . . . Number of areas on the casting that require repair due to
the method was abandoned and the heat director was inclusions or dirt.
responsible for keeping this gap between 4 and 5 inches Dirt Length . . . . Length of areas on the casting that require repair due to
for every mold. inclusions or dirt.
248 AFS Transactions