Central Composite Designs: When and How to Use Them in Six Sigma

Many Six Sigma teams start with simple experiments. They test a few factors. They compare high and low settings. That approach works early on. However, it breaks down once processes become more complex. At some point, linear models stop telling the full story. Curvature appears. Interactions grow stronger. Optimization becomes the real goal. That is where Central Composite Designs, or CCDs, become essential.

Central Composite Designs sit at the core of advanced DOE. They support Response Surface Methodology. More importantly, they help teams move from “improvement” to “optimization.” This article explains when to use CCDs, how they work, and how Six Sigma teams apply them in real projects.

What Is a Central Composite Design?

Central Composite Designs are a type of DOE used to model curvature. It extends a two-level factorial or fractional factorial design. Instead of stopping at linear effects, it adds points that estimate quadratic terms.

As a result, CCDs support second-order models. These models include squared terms and interaction terms. Therefore, teams can locate optimal settings rather than just directions for improvement.

A CCD contains three main components.

1. Factorial points. These points sit at the corners of the design space. They estimate main effects and interactions.
2. Axial points, also called star points. These points extend beyond the factorial levels. They estimate curvature.
3. Center points. These runs sit at the midpoint of all factors. They estimate pure error and detect curvature.

Together, these elements create a flexible and powerful design.

Why Central Composite Designs Matter in Six Sigma

Six Sigma projects often begin with screening designs. Teams identify critical Xs. They reduce noise. However, screening designs do not optimize processes.

At the Improve phase, teams need precision. They must understand curvature. They must balance tradeoffs. CCDs enable that shift.

Several benefits make CCDs especially valuable.

They reduce the number of runs compared to full three-level designs, they still estimate quadratic effects, and they provide strong prediction capability near the center of the design space.

In addition, CCDs align well with DMAIC logic. They build directly on earlier DOE work. Therefore, teams do not need to start over.

Most importantly, CCDs help teams find operating windows, not just single settings. That capability supports control and long-term stability.

When Should You Use a Central Composite Design?

CCD usage depends on timing and objectives. Using it too early wastes effort. Using it too late slows optimization.

Several clear signals indicate the right moment:

1. Screening experiments already identified key factors. Typically, this means two to five continuous factors remain.
2. Evidence of curvature exists. For example, center points in a factorial design show nonlinearity.
3. The project goal involves optimization. Cost, yield, cycle time, or performance often drive this need.
4. Factor ranges remain adjustable. CCDs require control over factor levels.
5. Resources allow additional runs. Although CCDs remain efficient, they still require more runs than simple designs.

When these conditions align, CCDs deliver strong value.

Central Composite Design vs Other Response Surface Designs

CCD is not the only response surface design. Box-Behnken designs also appear frequently. However, CCDs offer distinct advantages.

The table below highlights key differences.

Feature	Central Composite Design	Box-Behnken Design
Includes axial points	Yes	No
Requires corner points	Yes	No
Supports rotatability	Yes	Limited
Handles factor extremes	Strong	Moderate
Run size flexibility	High	Moderate

CCD flexibility stands out. Teams can customize axial distance. They can add center points. They can even build CCDs from existing factorial designs.

As a result, CCDs often become the default choice in Six Sigma optimization work.

Understanding the Structure of a CCD

To use CCDs effectively, teams must understand their structure. Each component serves a specific purpose.

Factorial Points

Factorial points come from a two-level factorial or fractional factorial design. They sit at coded levels of -1 and +1.

These points estimate main effects and interactions. They also anchor the design space.

For example, with three factors, a full factorial contains eight runs. Those runs form the backbone of the CCD.

Axial (Star) Points

Axial points extend beyond the factorial cube. They sit at ±α along each factor axis.

These points estimate quadratic effects. Without them, curvature remains invisible.

The value of α depends on design goals. For rotatable designs, α equals (number of factorial points)^(1/4).

Although axial points may seem extreme, they often fall within safe operating limits when scaled correctly.

Center Points

Center points appear at the midpoint of all factors. Teams usually replicate them.

These runs estimate pure error. They also test for curvature.

Additionally, center points stabilize the model. They improve prediction near normal operating conditions.

Rotatable vs Face-Centered CCDs

Not all CCDs look the same. Two common types dominate practice.

Rotatable CCDs

Rotatable designs maintain equal prediction variance at equal distances from the center. This property helps when direction matters more than exact location.

Rotatable CCDs use axial points outside the factorial region. Therefore, they require wider factor ranges.

Face-Centered CCDs

Face-centered CCDs place axial points at the same distance as factorial points. In coded units, α equals 1.

This structure simplifies experimentation. It also avoids extreme settings.

However, prediction variance varies with direction. That tradeoff may matter in some projects.

The table below summarizes the difference.

Aspect	Rotatable CCD	Face-Centered CCD
Axial distance (α)	> 1	1
Extreme factor levels	Wider	Limited
Prediction uniformity	Strong	Moderate
Ease of execution	Moderate	High

Choosing Factor Ranges for a CCD

Factor range selection determines CCD success. Poor ranges distort models. Good ranges reveal true behavior.

Teams should follow several guidelines:

1. Base ranges on prior screening experiments. Avoid guessing.
2. Confirm safety and feasibility at axial points. Always review limits with subject matter experts.
3. Center ranges around current operating conditions. This approach improves relevance.
4. Avoid overly narrow ranges. Curvature disappears if ranges shrink too much.
5. Scale factors carefully. Coded units help modeling, but real-world limits still matter.

Strong planning at this stage prevents wasted runs later.

How Many Runs Does a CCD Require?

CCD run size depends on factor count and design choices.

The general formula looks like this:

Total runs = factorial runs + axial runs + center points

For k factors:

• Factorial runs = 2^k (or fewer with fractional designs)
• Axial runs = 2k
• Center points = typically 4 to 6

The table below shows common examples.

Number of Factors	Typical Runs
2	13
3	20
4	30
5	42

Compared to full three-level designs, CCDs save significant effort.

Building a CCD Step by Step

Executing a CCD requires structure. Skipping steps creates confusion.

Step 1: Confirm Objectives

Start with a clear goal. Optimization must drive the experiment.

Define the response clearly. Yield, strength, or cycle time often work well.

Step 2: Select Factors

Limit factors to those that truly matter. Two to five works best.

Ensure factors remain continuous and controllable.

Step 3: Choose Design Type

Decide between rotatable and face-centered designs. Balance statistical rigor with practical limits.

Step 4: Determine Run Order

Randomize runs to protect against bias. Blocking may help if time or equipment changes matter.

Step 5: Execute and Measure

Run experiments carefully. Maintain consistent measurement systems.

Step 6: Analyze and Model

Fit a second-order regression model. Evaluate terms systematically.

Step 7: Optimize and Validate

Use contour plots and response surfaces. Confirm optimal settings with confirmation runs.

Example: Optimizing Cure Time and Temperature

Consider a manufacturing process involving thermal curing.

The team identified two critical factors:

• Cure temperature
• Cure time

Screening showed strong effects. Center points suggested curvature.

A CCD makes sense.

Design Setup

• Factors: 2
• Design type: Face-centered CCD
• Total runs: 13
• Response: Tensile strength

Results Summary

Analysis showed significant quadratic effects for temperature. Interaction effects also appeared.

Contour plots revealed a ridge rather than a single peak.

Outcome

The team selected a robust operating window. Strength increased by 12%. Variation dropped by 20%.

Most importantly, the process became easier to control.

Modeling and Interpreting CCD Results

CCD analysis focuses on second-order models.

The general form looks like this:

Y = β₀ + ΣβᵢXᵢ + ΣβᵢᵢXᵢ² + ΣβᵢⱼXᵢXⱼ

Each term adds insight.

Main effects show direction. Quadratic terms reveal curvature. Interaction terms explain tradeoffs.

Teams should follow a structured approach:

1. Review model adequacy. Check R², adjusted R², and residual plots.
2. Evaluate term significance. Remove weak terms cautiously.
3. Visualize results. Contour plots communicate insights faster than tables.
4. Confirm predictions with real runs.

Using Contour and Surface Plots Effectively

Plots bring CCD models to life. They help teams see tradeoffs instantly.

Contour plots show response levels across two factors. Surface plots add depth.

Best practices improve clarity.

• Fix non-plotted factors at meaningful levels
• Label axes clearly
• Avoid clutter
• Focus on operating windows, not just peaks

These visuals also support stakeholder alignment. Operators and leaders often trust plots more than equations.

Common Mistakes When Using CCDs

Even experienced teams make mistakes. Awareness prevents rework.

One common error involves using CCDs too early. Without screening, noise overwhelms results.

Another issue comes from poor factor ranges. Narrow ranges hide curvature. Wide ranges create infeasible conditions.

Teams also overfit models. Including every term reduces interpretability.

Ignoring confirmation runs causes trouble as well. Models need validation.

Finally, teams sometimes chase mathematical optima that lack robustness. Operating windows matter more than peaks.

Integrating CCDs into DMAIC

CCDs fit naturally into DMAIC.

DMAIC Phase	Relation to CCDs
Define	Teams clarify optimization goals
Measure	Teams confirm they have reliable data
Analyze	Teams build initial models
Improve	CCDs refine those models and locate optima
Control	Teams lock in settings and monitor stability

This alignment explains why CCDs appear so often in advanced Six Sigma projects.

Software and Practical Execution

Most statistical software, such as Minitab and JMP, supports CCDs. However, software does not replace thinking.

Teams must still choose ranges, validate assumptions, and interpret results.

Treat software as a calculator, not a decision-maker.

Templates, planning worksheets, and peer reviews all help maintain discipline.

When CCDs Are Not the Right Choice

CCD is powerful, but not universal.

Avoid CCDs when factors remain categorical. Also avoid them when resources remain extremely limited.

Processes with severe constraints may also struggle with axial points.

In those cases, alternative designs or sequential experimentation may work better.

Conclusion

Central Composite Designs represent a turning point in Six Sigma projects. They move teams beyond linear thinking. They unlock optimization.

When used correctly, CCDs deliver deep insight with reasonable effort. They support robust decision-making. They also align perfectly with DMAIC logic.

However, success depends on timing, planning, and interpretation. Teams must respect both statistics and process knowledge.

For practitioners ready to optimize rather than just improve, CCDs offer one of the most powerful tools in the Six Sigma toolbox.