SPC Analysis Workbench

What is SPC (Statistical Process Control)?

Statistical Process Control distinguishes common-cause variation (the inherent noise of a stable process) from special-cause variation (signals that something changed). Control charts plot process data against control limits computed from the data itself — typically at ±3 standard deviations from the centerline — so that only statistically unusual points trigger action.

This workbench selects the chart automatically from your subgroup size: individual values (n=1) use an I-MR chart pair, subgroups of 2–10 use Xbar-R, and larger subgroups use Xbar-S. The within-subgroup standard deviation estimated from the range or moving range (σ̂ = R̄/d₂ or MR̄/1.128) drives the control limits and the short-term capability indices Cp/Cpk, while the overall standard deviation of all data drives the performance indices Pp/Ppk.

Nelson rules 1–8 extend detection beyond simple limit violations: runs, trends, stratification, and oscillation patterns reveal non-random behavior even when every point is inside the limits. A process must be in statistical control before its capability indices are meaningful.

Formula: UCL/LCL = X̄ ± 3σ̂ (σ̂ = R̄/d₂ for Xbar-R, MR̄/1.128 for I-MR)

When to Use This Workbench

• During a pilot run or PPAP submission when you need control charts and capability indices (Cp/Cpk/Pp/Ppk) from the same data set in one pass
• For routine monitoring of a key characteristic where you key in or paste measurement series and need violations flagged automatically
• For single-value processes (batch chemistry, daily yield, slow production) where subgrouping is impossible — the I-MR chart handles n=1 correctly
• After a process change (tooling, material, parameters) to verify both stability (control chart) and capability (indices) before resuming production
• When investigating suspected instability — Nelson rules detect runs, trends, and stratification that simple limit checks miss

Common Mistakes to Avoid

• Using specification limits as control limits — control limits come from process variation, not from the drawing; confusing them leads to tampering or missed signals
• Judging capability from an unstable process — if the control chart shows violations, Cp/Cpk values are not trustworthy; remove special causes first
• Collecting too few subgroups — fewer than 20–25 subgroups make control limits and capability indices unreliable; treat early results as preliminary
• Mixing data from different machines, shifts, or material lots in one chart — stratified data inflates variation and hides signals; chart each stream separately
• Reacting to every point that wanders from the centerline — adjusting a stable process (tampering) adds variation instead of removing it

How to Interpret Results

• Points outside UCL/LCL or Nelson rule violations indicate special causes — investigate and remove them before trusting capability indices
• Cpk ≥ 1.33 meets the common industry requirement for ongoing production; Cpk ≥ 1.67 is typical for new or safety-critical processes
• A Ppk well below Cpk means the process drifts between subgroups — the short-term capability is not being realized over time
• Compare the histogram against the overlaid normal curve — strong skew or multiple peaks suggest the normality assumption behind the indices is violated
• Cp much higher than Cpk means the process is off-center — re-centering the mean is usually cheaper than reducing variation

Related Standards & References

• ISO 7870-2 — Shewhart control charts: constants (d₂, A₂, D₃/D₄, B₃/B₄) and control-limit methodology used for I-MR, Xbar-R, and Xbar-S charts
• AIAG SPC Reference Manual, 2nd Edition — automotive-industry methodology for control charting and capability studies (Cp/Cpk vs Pp/Ppk)
• Nelson, L.S. (1984), Journal of Quality Technology 16(4) — original publication of the eight Nelson rules implemented in this workbench
• ISO 22514 series — statistical methods for process capability and performance estimation
• Montgomery, D.C., Introduction to Statistical Quality Control — standard textbook treatment of rational subgrouping and chart selection

Frequently Asked Questions

Which control chart does the workbench choose?

It depends on the subgroup size you set: n=1 produces an Individuals & Moving-Range (I-MR) pair, n=2–10 produces Xbar-R (range-based), and n>10 produces Xbar-S (standard-deviation-based), following standard SPC practice.

What is the difference between Cp/Cpk and Pp/Ppk?

Cp/Cpk use the within-subgroup (short-term) standard deviation estimated from ranges, answering 'what is the process capable of at its best?'. Pp/Ppk use the overall standard deviation of all data, answering 'how did the process actually perform?'. A large gap between Cpk and Ppk indicates instability between subgroups.

What are the Nelson rules?

Eight statistical tests for non-random patterns on a control chart: points beyond 3σ (rule 1), long runs on one side of the centerline (rule 2), trends (rule 3), alternation (rule 4), and zone-based tests (rules 5–8). Each enabled rule is checked against every point; violations are listed with the rule number and point index.

Why are some of my data points excluded?

The preprocessing step drops empty/non-numeric cells and any value outside the optional cutoff bounds before analysis. The 'used' and 'excluded' counts show exactly how many points entered the calculation, so the charts are never silently distorted by sentinel values or typos.

Is this really free, or is it a limited trial for a paid product?

It's a complete, permanently free tool — every feature here (control charts, capability indices, pattern-rule checks, history, and export) runs entirely in your browser, with no account, no usage limit, and no paid tier to unlock. Unlike SPC/quality-monitoring platforms that gate control charts or capability analysis behind a subscription, this workbench has no server-side cost to recoup: the calculation runs 100% client-side, so there's no paid version waiting behind it.