Designing a Process Capability Analysis Workflow in Minitab

The Risk of Misleading Capability Indices

Engineers often rely on a single number—the Cpk—to determine if a manufacturing process is "capable." However, reporting a Cpk without first verifying statistical control and normality creates a high risk of false positives. A process can appear capable in a snapshot while being fundamentally unstable, leading to unexpected scrap rates and field failures.

The goal of a Capability Analysis is to determine if the natural spread of a process (the voice of the process) fits within the specification limits (the voice of the customer). The primary takeaway is that Cpk is only valid for stable, normally distributed processes; otherwise, you must use Ppk or non-normal transformations.

Requirements for Valid Analysis

Before running the Capability Analysis tool in Minitab, the dataset must meet three structural requirements:

• Defined Specification Limits: You must have a Lower Specification Limit (LSL) and/or an Upper Specification Limit (USL). These are engineering requirements, not process averages.
• Stability: The process must be in a state of statistical control. This means no special-cause variation (spikes or trends) is present.
• Normality: The data must follow a Gaussian (bell-shaped) distribution. Minitab's standard Cpk calculations assume this distribution to estimate the percentage of parts falling outside limits.

The Smallest Suitable Design for Capability Assessment

To avoid over-engineering the analysis, use this lean sequence of operations. This design ensures that each step validates the assumptions of the next.

Step 1: Stability Verification

Run an I-MR (Individual-Moving Range) Chart or an Xbar-R Chart. If the control chart shows points outside the control limits or non-random patterns, the process is unstable. In this state, Cpk is mathematically invalid because the process mean and variance are shifting.

Step 2: Normality Testing

Navigate to Stat > Basic Statistics > Normality Test. If the p-value is less than 0.05, the data is non-normal. You have two design choices here:

• Transformation: Use the Box-Cox or Johnson transformation to normalize the data.
• Non-Normal Analysis: Use the Stat > Quality Tools > Capability Analysis > Non-Normal path, which uses a different distribution model (e.g., Weibull).

Step 3: Capability Calculation

Run the Capability Analysis > Normal tool. Input your data column and the USL/LSL. Minitab will generate the Cpk (Process Capability Index) and Ppk (Process Performance Index).

Data Boundaries: Cpk vs. Ppk

A critical design decision is choosing which index to report. The difference lies in how they handle variation boundaries:

Metric	Variation Used	Engineering Context
Cpk	Within-subgroup (Short-term)	"What is the process capable of doing if we eliminate shifts?"
Ppk	Overall (Long-term)	"What did the customer actually receive over the last month?"

Operational Checks and Verification

To verify the software's output, perform a manual check on a small sample. For a process with only an Upper Specification Limit (USL):

Manual Cpk = (USL - Process Mean) / (3 * Sigma_within)

Run this calculation using the mean and standard deviation provided in the Minitab analysis summary. If the manual result differs significantly from the software's Cpk, verify that the correct subgroup size was defined in the input settings.

Diagnostic Decision Matrix

• Cpk > 1.33: Process is generally capable.
• Cpk < 1.00: Process is producing defects; immediate intervention required.
• Cpk > Ppk: Indicates the process is shifting or drifting over time (between-subgroup variation is high).

Failure Modes and Design Changes

The design of this analysis must change if the following conditions occur:

• Bimodal Distribution: If the histogram shows two peaks, you are likely measuring two different processes (e.g., two different machines) mixed together. You must split the data and run separate analyses.
• Single-Sided Specs: If only an LSL exists (e.g., minimum breaking strength), ensure the "Lower spec" field is filled and "Upper spec" is left blank to prevent Minitab from calculating a misleading average capability.
• High Non-Normality: If a Box-Cox transformation fails to normalize the data, move to a Non-Normal Capability analysis using the Weibull distribution.