7.1 Statistical process control (SPC): cₚ & cₚₖ

SPC is the financial difference between merely inspecting quality in at the end of the line (reactive waste) and actively building quality in at each step (proactive engineering). Inspection only tells you that a bad part was made; SPC indicates that the process is about to make a bad part. It is the practice of constantly listening to the statistical “heartbeat” of the manufacturing equipment. Ignoring these statistical trends means driving production blind, simply waiting for scrap to signal a problem.

Cₚ vs. cₚₖ: process capability

We use two distinct metrics to evaluate if a process can repeatedly meet the customer’s engineering requirements. It is important to clarify their distinct roles in customer reporting.

Cₚ (Process Potential): The Width

Definition: This metric compares the total width of the allowed tolerance (Specification Limits) to the natural measured variation of the machine process itself (6σ). It answers whether the distribution can theoretically fit within the limits.
The Logic: Should Cₚ fall below 1.0, the process variation is wider than the specification limits. Operator skill or retraining cannot resolve this. Instead, you must invest in better precision tooling, or discuss loosening the print tolerance with the design team. When Cₚ is 1.33 or greater, the process demonstrates the potential for stable mass production.

Cₚₖ (Process Capability): The Centering

Definition: This evaluates whether the process is currently centered on the nominal target, accounting for the real-world shift of the running mean from that target.
The Logic: When Cₚₖ drops below 1.0, parts of the distribution are falling out of spec, meaning scrap is being produced. When Cₚₖ equals Cₚ, the process is perfectly centered on nominal.

Pro-Tip: A high Cₚ paired with a low Cₚₖ means the equipment is highly precise (producing a tight grouping) but inaccurate (aimed off-target). This is often an easier issue to resolve by safely calibrating the machine offset back to nominal.

Control limits vs. specification limits

A common conceptual error on the shop floor is confusing calculated Control Limits with the actual Specification Limits.

Specification Limits (USL/LSL): These are explicitly defined by the Customer Print. Exceeding these limits means the part is non-conforming and cannot be shipped.
Control Limits (UCL/LCL): These are proactively defined by the Actual Process Data (typically calculated as ±3σ from the running mean). They serve as statistical guardrails.

The Guiding Rule:

Whenever a data point exceeds the expected Control Limit (UCL/LCL), the underlying process has changed. The line should be stopped for investigation. While the current part might still be functionally acceptable (within USL/LSL), the process itself is out of statistical control.

Conversely, if an operator selectively adjusts the machine simply because a part measures close to the Spec Limit while remaining inside the Control Limit, they are “Tampering.” This over-correction increases overall part variation and undermines predictability.

Reaction rules (the “nelson” logic)

Intervention should rely on defined, automated statistical triggers rather than operator intuition.

Stop & Fix Triggers:

The Outlier: A single data point falls outside the formally calculated Control Limits (±3σ).
The Trend: Seven consecutive data points move steadily in the same direction (Up or Down), indicating progressive tool wear or thermal drift.
The Shift: Seven consecutive data points are clustered on the same side of the Mean line. This signals that the process center has shifted, perhaps due to a new material lot or a fixture adjustment.

Final Checkout: Statistical process control (SPC): cₚ & cₚₖ

Control Point	Critical Engineering Requirement	Risk Avoided
Cₚₖ Target	Must be ≥ 1.33 (Standard), or ≥ 1.67 (Automotive/Safety Critical).	Scrap Generation.
Sampling Frequency	Subgroup size of exactly n=5 is standard for X-bar charts.	Insufficient data sensitivity.
Reaction Protocol	Line Stop required on any point > UCL/LCL.	Operating a drifted process.
Process Discipline	Avoid manually adjusting a running process that is “Statistically In Control.”	Destructive Tampering / Over-correction.
Limit Calculation	Recalculate Limits based on measured data every 50 subgroups.	Operating on stale, irrelevant control limits.