Types of Outputs

Outputs are classified according to our objective.

Objective Examples of Outputs
1. Target Value is Best
Achieve a • Part Dimension
target value with
minimum variation • Voltage
• Foam ILD

Target
2. Minimum Value is Best
Drive output • Cycle Time
toward zero • Part Shrinkage
• Warpage
0
3. Maximum Value is Best Drive output • Strength
upward • Durability

Johnson Controls, Inc. © 02.20.02 1 W4ReviewofImprovePhase.ppt

 Selecting a “Good” Output

To obtain the most useful information from a DOE

conducted to improve a process….

1. Use Variable Data!

2. Meaure something that relates to the basic function of
the process.

Johnson Controls, Inc. © 02.20.02 2 W4ReviewofImprovePhase.ppt

 Use Variable Data
Attribute data such as counting defects is an ineffective output
or a DOE because it….

 Provides insufficient detail for evaluating factor effects.

 Often leads to conclusions that are difficult to reproduce.
 Requires large sample sizes.

Variable Data overcomes these problems.

Johnson Controls, Inc. © 02.20.02 3 W4ReviewofImprovePhase.ppt

 Measure Outputs that Relate to Function
By using a measure of function, rather than a problem-related
measure, we can….
Minimize the chance for all problems to occur!!

Consider the molding example. The function of the process

is to create parts of a specific dimension. By achieving
minimum variation in part dimension, we have discovered a
way to uniformly distribute particles and stresses throughout
the part. When this happens, we minimize the occurrence
of voids, flash and other problems.

We do not have the same power to reduce problems

O TE when the problems themselves are measured as
N
the experimental output!!
Johnson Controls, Inc. © 02.20.02 4 W4ReviewofImprovePhase.ppt
 Strategy When “Target Value is Best”

Step 1: Find the factors that

affect variation. Use
these factors to
minimize variation.

Step 2: Find the factors that shift

the average (and don’t
affect variation). Use
these factors to adjust the
average output to the
desired target.
Target
Johnson Controls, Inc. © 02.20.02 5 W4ReviewofImprovePhase.ppt
 Strategy When “Minimum Value is Best”
Drive output
low

• Many transactional DOEs focus on minimizing processing

time or cycle time.

• The objective in this case is to find the factors that affect
the average output (time). Use these factors to drive the
average low.
• Minimizing variation in the output will also improve the
output when factors are found that strongly affect variation.
Johnson Controls, Inc. © 02.20.02 6 W4ReviewofImprovePhase.ppt
 Steps for Designing and Conducting DOE
1. Look at historical data and/or collect data to determine current process
capability.
2. Determine the objective of the experiment.
3. Determine what to measure as the output of the experiment.
4. Identify factors (control factors and noise factors) that could affect the output.
5. Determine the number of levels for each factor, and their actual values.
6. Select an experimental layout that will accommodate the selected factors and
levels and decide number of repetitions or replications.
7. Verify all measurement systems.
8. Plan and prepare the resources (people, materials, etc.) for conducting the
experiment. Should we randomize the runs? Create a test plan.
9. Conduct the experiment, assuring each unit is labeled according to the
experimental condition by which it was produced.
10. Measure the experimental units.
11. Analyze the data and identify strong factors.
12. Determine combination of factor levels that best achieves the objective
13. Run a confirmation experiment at this “optimal” combination.
14. Assure these best levels for strong factors are maintained over time by
implementing Standard Operating Procedures and visual controls.
15. Re-evaluate process capability.
Johnson Controls, Inc. © 02.20.02 7 W4ReviewofImprovePhase.ppt
 Keys to Successful Experiments
1. Good Output Measure
Whenever possible, use an output that directly relates to the function of the
process. When a measure of the problem or defect must be used, be sure to
use variable data.

2. Sound Experimental Design

No amount of data analysis can make up for a poorly designed experiment.
Carefully select output response, factors and levels and the DOE layout.

3. Careful Planning
To assure conditions can be controlled as stated in the experimental design, all
resources (people, materials, etc.) for conducting the experiment must be
prepared ahead of time.

4. Verified Measurement Systems

To assure the data is “good”, verify all measurement systems before conducting
the DOE.

5. Track the Experimental Units

Label each unit according to the experimental condition by which it was
produced. Otherwise, all information is lost.
Johnson Controls, Inc. © 02.20.02 8 W4ReviewofImprovePhase.ppt