PROCESS OPTIMIZATION AS A TOOL IN THE ANALYSIS
OF STEEL CASTING DEFECTS
Maynard Steel Casting Company
Milwaukee, WI 53217
PROCESS OPTIMIZATION AS A TOOL IN THE ANALYSIS
OF STEEL CASTING DEFECTS
Several technical papers are available on casting defects in the literature. However, quality costs
in steel foundries are still mainly due to the rework and scrap related to casting anomalies, which
are not acceptable to customers. In order to minimize the castings which do not meet the
customer acceptance specifications, it is not only necessary to identify the process parameters
related to the specific defects, but also it is necessary to identify the levels of these parameters to
produce acceptable castings. Metal casting process has several sub-processes, which in turn
have a number of process variables, which could influence the occurrence of defects in castings.
Conventional statistical techniques with design of experiments involves too much work for the
foundries to identify the process variables and their levels responsible for the defects. The use of
a process optimization tool which uses foundry production data that can be collected on the
castings on a regular basis in the identification of the process variables and their levels will be
presented and discussed in this paper. This is useful in converting production data into
actionable information that leads to minimization of quality costs in steel foundries.
Process optimization is the identification and control of the input process parameters (Factors) to
achieve the desired output (Response) in any process. Metal casting process is a complex
process with several sub-processes, such as patternmaking, mold and coremaking, melting and
pouring, heat treatment and cleaning and finishing. Six Sigma methodologies have been
attempted in steel foundries to minimize the casting defects and improve profitability. Six Sigma
uses DMAIC methodology to improve the processes. Six sigma heavily focuses on statistical
analysis as it is data driven and is a methodical approach that drives the process improvements
through statistical measurements and analyses. In view of the large number of factors that are
responsible for the casting defects, the general statistical approach is not always the best. An
alternate and more elegant pattern recognition approach is found to be appropriate for the metal
casting related issues. It is suggested that the foundries follow the Six Sigma methodology with
the exception that a pattern recognition process optimizer is used instead of the conventional
Design of Experiments.
Figure 1 shows the process map of the casting process. It is a SIPOC diagram. On the left hand
side, all the inputs which are Xs are presented and on the right hand side all the outputs are
presented which are Ys. These Ys include the casting defects.
Casting defects which result in rework and scrap are a major issue in steel foundries. It is
estimated that about 5 to 10% of the revenues of the foundries are lost in internal and external
failure costs. For a 100 million dollar company a reduction of even 1% of these costs will result in
a savings of 1 million dollars per year.
In the Define phase of Six Sigma methodology, a project charter is made with specific
measurable goals. This is the critical phase of the methodology where the appropriate goal is
specified based on the resources available to successfully complete the project. There will not be
any change in the Define phase in the present suggested route. The project goal in the project
charter specifies the big Y that will be optimized.
Figure 2 shows an example of the CTQ flow diagram. This is commonly used to identify the Ys
which are specified in the project charter.
In the Measure phase a tool known as Cause and Effect diagram is used to identify all the
factors that are responsible for the casting defects. An example of the Cause and Effect diagram
is shown in figure 3. Here the effect ‘Inclusions’ is the Y and all the causes are Xs.
Essentially what matters is the recognition of all the factors that are responsible or related to the
appropriate casting defects. This can be represented in the form of a relation:
Y is a function of various Xs. This is shown in figure 4 as Y = F(X)
Cause and Effect matrix shown in figure 5 is another useful tool in identifying all the relevant Xs
related to the Y of interest in the project. As can be seen from this diagram there are nearly 25
Xs that are related to a single Y, namely Inclusions.
Figure 6 shows an example of Failure Mode and Effects Analysis diagram which filters the
potential Xs into critical Xs.
In the conventional Six Sigma approach all the potential Xs are filtered into critical Xs as shown in
figure 7 in the various phases of the DMAIC methodology.
In the Improve phase of DMAIC methodology, Design of Experiments are extensively used to
identify the factors and their levels that are related to the response variable namely the casting
Limitations of DOE as applied to foundry processes.
1. In the DOE, the number of experiments needed depends on the number of factors. In the
2K design, each factor is varied over two levels and the number of experiments needed
are 2 where K is the number of factors. For 3 factors, 8 experiments are needed and
for 4 factors, 16 experiments are needed. As the number of factors increase, the number
of experiments needed increase exponentially. As can be seen in figure 1, where there
are a large number of factors, it becomes very cumbersome to carry out design of
experiments. Even if fractional factorial designs or Taguchi design of experiments are
used , the number of experiments needed become very large.
2. In the DOE, there is a need to carry out controlled experiments to collect the required
data and it could interrupt regular production.
3. In the DOE, the level of factors have to be with a considerable difference in order to have
4. The DOE assumes known distributions to the unknown foundry processes, as such the
results could be biased.
5. The DOE requires people with a reasonable expertise in the use to statistical techniques
to design the experiments and interpret the results.
Y IS A FUNCTION OF X:
In the analysis of casting defects, the casting process is optimized to minimize the defects. The
concept of Y = F (X) is used in the analysis.
Ys: Responses :
These are the manifestation of the process. These are measures of the output of the process.
Casting Defects are appropriate responses which need to be minimized in the optimization of
Examples: Inclusions, Misrun, Shrinkage, Cracks etc
Xs : Factors :
Process variables which have an effect on Y
These are the parameters which control the output of a process
Examples: Pouring Temp, Pouring Time, Compactability, Gating Ratio, Operator, Shift, Design
It is the identification and control of the input process Parameters (Factors) to achieve the desired
output (Response) in any process
PROCESS OPTIZATION SOFTWARE (MetaCause)
MetaCause Process optimization software (www.metacause.com) is a pattern recognition
optimization software based on three powerful concepts (1)namely:
1. Receiver Operator Characteristics (ROC)
2. Relationship Hyper-Surface
3. Interactions and Impossible Probability
In general, the relationships between the factors and responses can be represented
schematically as shown in figure 8. The various factors and responses are inter-related as shown
in this diagram with connecting lines.
The process optimisation software recognizes the patterns and identifies their significance as
shown in figure 9. The thick line represents a positive effect of the factor on the response
variable. The dotted line represents a negative effect of the factor on the response variable and
the dotted line with dots indicates that the factor has no effect on the response variable. The
thickness of these lines indicates the weightage of the factor over the response variable.
Figure 10 shows a typical user friendly output of the process optimisation software to interpret the
results of the analysis.
ADVANTAGES OF PROCESS OPTIMIZATION SOFTWARE:
1. It uses the actual production data that is collected during the operation of the process. It uses
all the factors and all the responses, rather than the filtered factors in conventional statistical
2. It can handle up to 50 factors and 25 responses at a time which is not practical with the
existing statistical tools.
3. The results are based on sound scientific principles based on scientific concepts of Receiver
Operator Characteristic, Relationship Hyper-Surface and Interactions and Impossible Probability.
Failure of test results for mechanical properties is one of the issues in steel foundries. An attempt
is made to determine the relationship between the chemistry of the heats to the mechanical
properties of the test blocks. The data collected is shown in figure 11. The various responses
identified in this analysis include, UTS, YS, %El, %RA, Charpys at –40F, Charpys at 70F,
Fracture at surface and Fracture at the center. The various factors considered include, the
chemistry of the heat with elements: C, Mn, Si, S, P, Ni, Cr, Mo, Al, Cu, Ca, Ti and also Hardness
of the test block in BHN. 20 observations were made in this case study relating the various
factors to the responses.
The results of the analysis are shown in figures 12 to 14. Figure 12 shows the importance
weighting interpretation. 100% refers to theoretical maximum, 65% and above indicates very
important process setting. Generally this is easily recognizable by the experts by viewing the
data itself.55% refers to highly influential setting and 50% refers to moderately important setting.
The ranges between 50 and 55% is difficult to be identified by the expert by viewing the raw
data.45% is the basic threshold for the important setting and 40% refers to noisy data and has no
relation to the response variable, by itself but may be significant as interaction variable with
another factor which needs to be examined further. This figure also shows the various names of
response variables, namely the mechanical properties including the charpy impact properties.
Figure 13 shows the optimal process setting for the response variable ‘Charpy value” –
CVN1.This slide has the results of the analysis showing the importance weighting of various
factors on the response variable CVN1. The software also gives the importance weight of the
factors for all the response variables chosen in the analysis. This result shows that low settings
of Mo and C have an importance weighting of over 50% and is significant to control to achieve the
optimum results. The other factors are not as significant compared to Mo and C.
Figure 14 shows the process settings to avoid. This slide indicates that low setting of Al and high
setting of C should be avoided to achieve the optimum results. The other factors listed in this
slide are not significant.
Figure 15 shows the factors that have no significant effect on the response variable. The process
optimization software gives three types of results, namely: the factors with optimal process
setting, the factors with process settings to avoid and the factors that have no significant effect on
the response variable of interest.
Using the above information the factors and the their settings are selected to achieve the
optimum performance of the response variables.
Recommended Steps in Casting Defect Analysis:
1. Use Six Sigma DMAIC Methodology.
2. In the Define phase, create a Project Charter with clearly idenfied measurable goal. It is
desirable to indicate a financial saving of at least $3000.00 per month continuously after
the project is successfully completed. It is desirable not select project of the type ”Boil the
Ocean” and with a large time frame. It is desirable to identify projects that have a time
frame of 6 weeks. It is desirable keep the project goal to one defect type or one part with
all possible defects. This will be the big Y for the project. Pareto Charts with different
levels are extensively used to identify the relevant projects.
3. In the Measure phase, benchmark the status of the big Y before the project commences.
Quantification of the defect is an important step in the measure phase. It is desirable
develop methods to quantify the defect. Use Cause and Effect Diagram to identify all the
factors Xs that can be related to the response variable Y. The Xs should have a metric
that is measurable. If necessary, Cause and Effect matrix can be used to filter a few of
4. In the Analyze phase use MetaCause process optimization software to identify the factors
with optimal settings, with negative effect and with no effect on the Y.
5. In the Improve phase validate the results based on the results obtained in the Analyze
phase and calculate the financial savings.
6. In the Control phase, develop a control plan and reaction plan and give to the process
Metal casting process is a complex process with several sub-processes. Six Sigma
methodologies are commonly used to optimize the process and minimize casting defects.
However, the conventional statistical tools available today are not adequate to be effective in
analyzing the casting defects and optimize the processes to minimize the impact on cost of
quality. The reason for these include: the statistical techniques assume known distributions to the
unknown foundry processes; the need for specially designed experiments; the need for carrying
out a very large number of experiments in view of the large number of factors; the need to carry
out specially designed experiments on a limited number of castings and the need to filter the
potential factors into a manageable number of factors. A process optimization software based on
pattern recognition is found to be suitable to optimize foundry processes and to minimize the
casting defects. A six sigma methodology is presented to address the issue of casting defects in
steel foundries with the exception that a pattern recognition process optimization software is used
instead of DOE techniques.