7.3 Statistical Process Control (SPC), Control Charts (X-bar & R), & Process Capability (Cp/Cpk)

Key Takeaways

  • Statistical process control (SPC) uses the Xˉ\bar{X} chart for subgroup averages and the RR chart for subgroup ranges.

  • Control limits come from process data. Specification limits come from product requirements.

  • For a stable process with two specification limits, CpC_p compares width and CpkC_{pk} accounts for the distance from the mean to the nearer limit.

  • If CpkC_{pk} is lower than CpC_p, the mean is off the specification midpoint. The gap alone does not establish when or why the mean moved.

  • For an aerospace job that invokes AS9102C, Form 3 records each design characteristic and its verification result.

Last updated: November 2026

7.3 Statistical Process Control (SPC), Control Charts (Xˉ\bar{X} & RR), & Process Capability (Cp/CpkC_p/C_{pk})

Statistical Process Control (SPC) uses measurements in production order to study how a machining process behaves. For example, subgroup averages can reveal a change in bore diameter even while all the most recent bores are within the drawing limits. Control-chart results are one input to the reaction plan for the job. This section provides supplemental shop-quality context for the NIMS inspection topics.

Fundamentals of SPC & Variation Types

No manufacturing process produces two parts that are identical down to the atomic level. Variation is inherent in every machining operation.

                   PROCESS VARIATION TYPES
                              |
         +--------------------+--------------------+
         |                                         |
  COMMON CAUSES                             SPECIAL CAUSES
  (Chance Variation)                        (Assignable Causes)
  - Minor ambient temp shifts               - Dull / broken insert
  - Microscopic material variation          - Incorrect work offset
  - Natural machine vibration               - Loose vise clamping
  ---> Stable & Predictable                 ---> Unstable & Out of Control

Common Cause Variation (Chance Variation)

  • Inherent, natural noise within a process operating normally.
  • Examples: Microscopic fluctuations in raw material bar stock hardness, minor room air temperature changes, or background hydraulic pump vibration.
  • Statistical Status: When a process exhibits only common cause variation, it is in a state of statistical control.

Special Cause Variation (Assignable Variation)

  • Variation associated with a particular event or change in conditions.
  • Examples: A chipped carbide insert edge, a loose spindle drawbar, a wrong tool offset entry, or an unannounced batch change in coolant concentration.
  • Action Required: A control-chart signal prompts investigation under the job’s reaction plan. Check the measurement and affected parts, and look for a specific change in the setup, tool or process before making an adjustment.

Sampling Plans & Subgroups

SPC data is collected in rational subgroups—small samples of sequential parts (typically subgroup size n=4n = 4 or n=5n = 5) taken at fixed time intervals (e.g., 5 parts every 2 hours). Sampling within a brief window minimizes within-subgroup variation while exposing process shifts over time.


Control Charts: Xˉ\bar{X} (X-bar) and RR Charts

Control charts track two fundamental parameters over time:

  1. Xˉ\bar{X} Chart (Sample Mean): Monitors process centering (location of the average dimension).
  2. RR Chart (Sample Range): Monitors process dispersion (spread or variability within the subgroup: R=Xmax⁡−Xmin⁡R = X_{\max} - X_{\min}).

Statistical Calculations & Formulas

Given kk subgroups of size nn:

  1. Subgroup Mean (Xˉ\bar{X}): Xˉ=∑i=1nXin\bar{X} = \frac{\sum_{i=1}^n X_i}{n}
  2. Grand Mean (Xˉˉ\bar{\bar{X}}): Xˉˉ=∑j=1kXˉjk\bar{\bar{X}} = \frac{\sum_{j=1}^k \bar{X}_j}{k}
  3. Subgroup Range (RR): R=Xmax⁡−Xmin⁡R = X_{\max} - X_{\min}
  4. Average Range (Rˉ\bar{R}): Rˉ=∑j=1kRjk\bar{R} = \frac{\sum_{j=1}^k R_j}{k}

Control Limit Formulas

Control limits use three estimated standard deviations of the charted statistic. For the mean chart this includes the subgroup-size effect. A range has its own distribution and cannot be read as an individual dimension or as a normal distribution.

X-bar Chart Limits:UCLXˉ=Xˉˉ+A2Rˉ,LCLXˉ=Xˉˉ−A2Rˉ\text{\textbf{X-bar Chart Limits:}} \quad \text{UCL}_{\bar{X}} = \bar{\bar{X}} + A_2 \bar{R}, \qquad \text{LCL}_{\bar{X}} = \bar{\bar{X}} - A_2 \bar{R}

R Chart Limits:UCLR=D4Rˉ,LCLR=D3Rˉ\text{\textbf{R Chart Limits:}} \qquad \text{UCL}_R = D_4 \bar{R}, \qquad \text{LCL}_R = D_3 \bar{R}

SPC Control Chart Constants Table

Subgroup Size (nn)Mean Factor (A2A_2)Range Factor (D3D_3)Range Factor (D4D_4)Std Dev Factor (d2d_2)
21.88003.2671.128
31.02302.5751.693
40.72902.2822.059
50.57702.1152.326
60.48302.0042.534

Worked Example: Control Chart Limit Calculation

Problem: A CNC turning cell inspects a pin diameter. k=25k = 25 subgroups of size n=5n = 5 yield a grand mean Xˉˉ=1.2500′′\bar{\bar{X}} = 1.2500'' and an average range Rˉ=0.0008′′\bar{R} = 0.0008''. Calculate the upper and lower control limits.

  1. Lookup constants for n=5n = 5: A2=0.577A_2 = 0.577, D3=0D_3 = 0, D4=2.115D_4 = 2.115.
  2. Calculate Xˉ\bar{X} Chart limits: UCLXˉ=1.2500′′+(0.577×0.0008′′)≈1.25046′′\text{UCL}_{\bar{X}} = 1.2500'' + (0.577 \times 0.0008'') \approx 1.25046'' LCLXˉ=1.2500′′−(0.577×0.0008′′)≈1.24954′′\text{LCL}_{\bar{X}} = 1.2500'' - (0.577 \times 0.0008'') \approx 1.24954''
  3. Calculate RR Chart limits: UCLR=2.115×0.0008′′≈0.00169′′\text{UCL}_R = 2.115 \times 0.0008'' \approx 0.00169'' LCLR=0×0.0008′′=0.00000′′\text{LCL}_R = 0 \times 0.0008'' = 0.00000''

Process Capability Analysis (CpC_p & CpkC_{pk})

A process can be in statistical control (predictable) yet still produce non-conforming parts if the tolerance limits are tighter than the process spread.

CRITICAL DISTINCTION:

  • Specification Limits (USL / LSL): Set by engineering design drawings ("Voice of the Customer").
  • Control Limits (UCL / LCL): Calculated from statistical process data ("Voice of the Process").
                 LSL              NOMINAL             USL
                  |                  |                  |
  Tolerance Spec: |<----------------------------------->|
                  |                  |                  |
  Process Spread:        |******  MEAN  ******|
                         |<------------------>|
                                6 Sigma

Process Capability Index (CpC_p)

CpC_p measures potential process capability assuming the process mean is centered perfectly between USL and LSL. It compares total allowable tolerance to total process spread (6σ6\sigma):

Cp=USL−LSL6σC_p = \frac{\text{USL} - \text{LSL}}{6\sigma} With rational subgroups, estimate within-subgroup standard deviation as σ^=Rˉ/d2\hat{\sigma} = \bar{R} / d_2. Establish stability before interpreting capability.

Process Capability Ratio (CpkC_{pk})

CpkC_{pk} accounts for process off-centering. It uses the smaller of the two signed margins from the mean to each specification limit:

Cpk=min⁡(Cpu,Cpl)=min⁡(USL−μ3σ,μ−LSL3σ)C_{pk} = \min \left( C_{pu}, C_{pl} \right) = \min \left( \frac{\text{USL} - \mu}{3\sigma}, \frac{\mu - \text{LSL}}{3\sigma} \right)

  • If Cp=CpkC_p = C_{pk}, the process mean is at the midpoint of the two specification limits. The midpoint may differ from the drawing’s nominal value.
  • If Cpk<CpC_{pk} < C_p, the mean is closer to one limit. Even a process with a spread narrow enough to fit between the limits can fall short through off-centering.
  • If Cpk<1.0C_{pk} < 1.0, the smaller signed margin is less than three standard deviations, including a negative margin if the mean lies beyond a spec limit. This alone does not show that the full process width exceeds the tolerance band.

Expected Rejects in a Centered, Normal Process

For stable, normally distributed output centered between two specification limits, Cp=CpkC_p=C_{pk}. In that model the reject rate is the sum of the two normal-distribution tails beyond the limits. The table shows how the rate changes with the distance to each limit.

Distance from mean to each limitCapability, roundedApproximate rejects
3σ3\sigma1.001.002,700 parts per million
4σ4\sigma1.331.3363 parts per million
5σ5\sigma1.671.670.57 parts per million
6σ6\sigma2.002.000.002 parts per million

On an off-center process, the two tails are different, so the nearer-limit index alone does not determine their sum. Assess whether the model and the measurement system are appropriate and compare the resulting index with the criterion in the applicable plan or contract.

Worked Example: CpC_p and CpkC_{pk} Calculation

Problem: A ground shaft journal print specifies a diameter of 0.7500′′±0.0015′′0.7500'' \pm 0.0015'' (USL =0.7515′′= 0.7515'', LSL =0.7485′′= 0.7485''). A stability check and subgroup measurements support a mean μ=0.7506′′\mu = 0.7506'' and standard deviation σ=0.0003′′\sigma = 0.0003''. Calculate CpC_p and CpkC_{pk}.

  1. Calculate total tolerance spread: USL−LSL=0.7515′′−0.7485′′=0.0030′′\text{USL} - \text{LSL} = 0.7515'' - 0.7485'' = 0.0030''.
  2. Calculate CpC_p: Cp=0.0030′′6×0.0003′′=0.0030′′0.0018′′≈1.67C_p = \frac{0.0030''}{6 \times 0.0003''} = \frac{0.0030''}{0.0018''} \approx 1.67
  3. Calculate upper and lower capability indices: Cpu=0.7515′′−0.7506′′3×0.0003′′=0.0009′′0.0009′′=1.00C_{pu} = \frac{0.7515'' - 0.7506''}{3 \times 0.0003''} = \frac{0.0009''}{0.0009''} = 1.00 Cpl=0.7506′′−0.7485′′3×0.0003′′=0.0021′′0.0009′′≈2.33C_{pl} = \frac{0.7506'' - 0.7485''}{3 \times 0.0003''} = \frac{0.0021''}{0.0009''} \approx 2.33
  4. Determine CpkC_{pk}: Cpk=min⁡(1.00,2.33)=1.00C_{pk} = \min(1.00, 2.33) = 1.00

Diagnosis: The mean is 0.0006 in toward the upper limit. If the process can be recentered at 0.7500 in with its spread unchanged, both limits will be 0.0015 in away and CpkC_{pk} will be about 1.67. That is a prediction, to be confirmed with fresh measurements. If an offset change has been authorized, check whether the specific control’s wear value acts on radius or diameter and verify its sign before entering it. Check the first adjusted piece and re-establish stability before re-estimating capability.

First Article Inspection in Aerospace

When an aerospace job invokes AS9102C, first article inspection (FAI) documents an evaluation of the applicable production process against the design requirements. It complements the customer’s requirements. An acceptable first article and a statistical study answer different questions. The first records verification of the applicable design characteristics. A capability study uses repeated measurements to assess the stable process relative to tolerances.

AS9102C formRecords
Form 1 — Part Number AccountabilityPart identity, drawing and revision, related part numbers where applicable, and whether the evaluation is full or partial
Form 2 — Product AccountabilityMaterial, special process and functional test evidence
Form 3 — Characteristic Accountability, Verification, and Compatibility EvaluationEach characteristic identifier and reference, its requirement and result, and designed or qualified tooling or a nonconformance reference when applicable

For example, link a hole’s drawing callout to its characteristic number on Form 3, record the complete dimension or geometric requirement, and record the measured or verified result. If a change might affect conformance, use the invoked standard and customer requirements to determine what to reassess and document the scope and reason.

See NIST’s control-chart formulas, NIST’s process capability guidance, and IAQG’s AS9102C forms.

Loading diagram...
Process Capability (Cpk) Distribution vs Tolerance Limits
Test Your Knowledge

For a rational subgroup size of n = 5, what factor A2 is used to calculate the Upper Control Limit (UCL) for an X-bar control chart?

A

0.729

B

0.577

C

2.114

D

1.330

Test Your Knowledge

For a stable process with two specification limits, the calculated Cp is 1.67 and Cpk is 0.95. What can you conclude from these values?

A

The process variation spread is too wide for the tolerance band and requires a new machine tool

B

The process has zero special cause variation and is in ideal statistical control

C

The estimated six-standard-deviation width is narrow enough to fit between the limits, but the mean is off the specification midpoint.

D

The measurement instruments used to record sample data are out of calibration

Test Your Knowledge

A stable process has normally distributed output centered between two specification limits. Each limit is four standard deviations from the mean. What approximate total reject rate does that model predict?

A

2,700 parts per million

B

0.57 parts per million

C

0.002 parts per million

D

63 parts per million

Test Your Knowledge

On a job that invokes AS9102C, which first article inspection form links the drawing characteristic and its requirement to its verification result?

A

Form 3: Characteristic Accountability, Verification, and Compatibility Evaluation

B

Form 1: Part Number Accountability

C

Form 2: Product Accountability

D

Form 4: Machine Tool Registration

Sections you finish are checked off in the contents.