📈 STATISTICAL PROCESS CONTROL · CP / CPK / CPM

SPC & Quality Control

Deep hole drilling is a blind, serial process — by the time you measure a finished hole, the tool may already be drifting and making more bad ones. Statistical process control turns “drill and hope” into a predictable, data-driven operation: control charts catch shifts before they become non-conforming holes, and Cp/Cpk/Cpm tell you exactly how much room your process has inside the tolerance band.

1.33 / 1.67Cp / CpkAcceptance thresholds
7Chart typesX-bar R → u chart
30–125PiecesCapability study size
~66 ppmAt Cpk 1.33Expected defect rate

Why SPC Matters for Deep Holes

Most machining is measured after the cut. Deep hole drilling is different — four characteristics make SPC unusually valuable here and force a tailored approach.

🚫
Blind process

The cutting zone is metres away at the bottom of the bore. Quality deviations must be inferred from measurements of finished holes — there is no in-process visual inspection.

🕑
Serial processing

A single hole takes minutes to hours to complete. By the time you measure the result, the tool may already have degraded and produced more bad holes.

💰
High cost of failure

A non-conforming bore can scrap a workpiece worth thousands of dollars. SPC’s early-warning capability is worth more here than in almost any other machining operation.

📦
Small batches

Aerospace and medical deep hole work is often 10–100 parts per run. That rules out classic high-volume charting and forces I-MR and short-run chart strategies.

💡 The rule that governs everything: stability comes before capability. Cp/Cpk are only meaningful once the process is in statistical control — no special causes. Published case studies consistently follow this order: establish control, remove assignable causes, then quantify capability. A connecting-rod machining study that did exactly this moved Cpk from 0.12 to 1.37 and Cp from 0.12 to 1.72 by attacking root causes (machine maintenance, tool-wear compensation, gauge verification) before re-measuring.

Key Quality Characteristics to Monitor

CharacteristicTypical SpecificationMeasurement MethodProcess Sensitivity
Bore diameter±0.02–0.10 mmAir gauge, CMM, bore micrometerTool wear, coolant pressure, material hardness variation
Straightness0.02–0.10 mm/mLaser bore alignment, straightness mandrelTool geometry, feed rate, material anisotropy, guide bushing condition
Surface roughness (Ra)0.4–6.0 µmProfilometer, surface comparatorTool edge condition, coolant type, cutting speed, vibration
Roundness0.01–0.05 mmRoundness gauge, CMMSpindle bearing condition, clamping force, tool imbalance
Position / location±0.10–0.50 mmCMM, coordinate measurementPilot hole accuracy, spindle alignment, thermal drift
⚠️ Measure diameter twice, not once: a single diameter reading per bore masks roundness error and tool-wear signals. Record high/low values (or measure at two angular positions) so the roundness component shows up in the subgroup range — otherwise a “stable” chart hides a lobed hole.

Cp, Cpk, Cpm & Ppk — What They Mean

Cp and Cpk measure spread and centering relative to the tolerance band; Cpm adds a penalty for missing the target value; Ppk reflects long-term actual performance including drift.

The formulas

IndexFormulaWhat it measuresWhen to use
Cp(USL − LSL) / 6σPotential capability — spread vs tolerance, assuming a centred processFirst check: is the spread even close to the band?
Cpkmin[(USL − μ) / 3σ, (μ − LSL) / 3σ]Actual capability — spread AND centeringPrimary acceptance index for a new or qualified process
Cpm(USL − LSL) / 6√(σ² + (μ − T)²)Target-based capability — penalises off-target runsWhen the target diameter matters as much as the tolerance
Ppkmin[(USL − μ), (μ − LSL)] / 3σoverallLong-term performance over all data, incl. between-subgroup driftProduction-phase verification and PPAP sign-off

Cpk vs Ppk — short-term potential vs long-term performance

CpkPpk
Sigma estimateWithin-subgroup (short-term, “potential”)Overall (long-term, “actual”)
CapturesVariation inside a stable windowTool wear, tool changes, setups, operators, material lots, thermal drift
Typical useProcess qualification, with control chartsProduction performance over real time
Common threshold≥ 1.33≥ 1.67
💡 Reading the pair: if Ppk is much lower than Cpk, the process drifts between subgroups — eliminate the assignable causes (tool wear, setups, coolant temperature) before chasing the tolerance. If Cpk ≈ Ppk, variation is almost all common cause and you are looking at genuine machine capability. Ppk higher than Cpk usually signals data mixed from different process states.

Acceptance criteria & expected defect rates

Cp / CpkRatingAction requiredExpected ppm (two-sided)
< 1.00Not capableProcess redesign — specifications cannot be met> 2,700
1.00 – 1.33MarginalClose monitoring; 100% inspection recommended; improvement needed2,700 → ~66
1.33 – 1.67CapableStandard SPC monitoring sufficient; acceptable for most applications~66 → ~0.55
1.67 – 2.00ExcellentReduced inspection possible; process well-controlled~0.55 → ~0.002
≥ 2.00World-classMinimal inspection needed; highly predictable< 0.002
⚠️ Know your customer’s bar: automotive PPAP commonly requires Cpk ≥ 1.67 for production-phase sign-off, with 1.33 treated as “currently acceptable but requires improvement.” Aerospace and safety-critical bores often demand ≥ 1.67 and sometimes 2.00. Confirm the requirement before quoting capability.

Choosing the Right Chart

Chart choice follows data structure first, then process behaviour. Variables (measured) charts handle bore diameter, straightness and roughness; attribute charts handle counts such as non-conforming holes or defects per hole.

ChartDataSubgroupDetectsBest for deep hole drilling
X-bar & RContinuous2–10, equal sizeShift in mean; change in spreadMedium/high-volume bores sampled in runs of 3–5
X-bar & sContinuous> 8–10, or unequal sizesSame, with a better sigma estimate at large nLarge subgroups fed by automated gaging
I-MRContinuous1 (each hole is its own subgroup)Point-to-point drift and tool wearSmall batches, low-volume and prototype bores
p / npAttribute (fraction / count defective)Fixed or variable nChange in non-conformance rateScrap / rework rate of drilled parts
u / cAttribute (defects per unit / count)Fixed or variable areaChange in defect densityMultiple defects per hole (chatter marks, recuts)
💡 Rule of thumb: for subgroup sizes up to about 8–10, estimate sigma from the range (R); for larger or unequal subgroups switch to the standard deviation (s). I-MR charts need approximately normal data — X-bar charts are robust to non-normality through the central limit theorem, I-MR is not. When you run a family of similar bore sizes in short runs, plot deviation-from-nominal (DNOM) so several diameters share one chart.
🕑
Subgroups of 3–5→ X-bar & R
📦
Small batches→ I-MR
📈
Large subgroups→ X-bar & s
⚠️
Defect counts→ p / np / u / c

Setting Control Limits from the Process

Control limits are the “voice of the process” — they must be calculated from process data, never from the drawing. Standard procedure:

  1. Collect 20–30 subgroups of data from the process running under stable conditions
  2. Calculate the grand mean (X-double-bar) and the average range (R-bar)
  3. Calculate the upper and lower control limits (UCL, LCL) using the standard A₂, D₃, D₄ factors
  4. Plot the limits on the chart — they represent the natural process variation
  5. Monitor ongoing production against these limits; recalculate after any significant process change
Subgroup size nA₂D₃D₄
21.88003.267
31.02302.574
40.72902.282
50.57702.114
⚠️ Voice of the customer vs voice of the process: specification limits define what the customer will accept; control limits define what the process naturally produces. Never plot specification limits as control limits and never calculate limits from the tolerance. A process inside specification but outside control limits is unstable and will eventually make non-conforming parts — and a stable process can still be out of spec, which is a capability problem, not something adjusting the machine fixes.

Running a Capability Study

A capability study proves a deep hole process can hold tolerance before you commit production. Work through it in this order:

1
Define critical characteristics

Identify which hole characteristics are critical to function. For most deep holes, diameter and straightness are primary; roughness and roundness become primary for hydraulic and bearing surfaces.

2
Verify the measurement system

Run a gauge R&R (GR&R) and calibrate before collecting anything. A gauge that consumes a large share of tolerance sinks the whole study.

3
Establish statistical control

Collect 20–30 subgroups under stable conditions, compute X-double-bar, R-bar and control limits, and remove every special cause until the chart is in control.

4
Collect the capability sample

Measure 30–125 consecutive parts produced under repeatability conditions with no tool change or setup break mid-study.

5
Check the distribution

Histogram plus a normality test. For non-normal data apply a transform (Box-Cox or Johnson) or use the Clements percentile method before computing indices.

6
Compute the indices

Calculate Cp, Cpk and Cpm against the specification; calculate Ppk for the long-term view across all data collected.

7
Decide & act

Compare against acceptance criteria (1.33 capable, 1.67 critical). If marginal, attack variation and tool path before re-quoting capability.

8
Hand off to monitoring

Deploy ongoing control charts, recalculate limits after any change — new tool supplier, material grade, machine maintenance, or coolant change.

Sampling Frequency, Sample Size & Measurement

Sampling frequency

Production VolumeSampling FrequencySubgroup Size
High volume (>1000 parts/day)Every 20th–50th hole3–5
Medium volume (100–1000/day)Every 10th–20th hole3–5
Low volume (<100/day)Every hole (100% inspection)1 (I-MR chart)
Prototype / first article100% inspectionN/A — characterise the process

30-piece vs 125-piece studies

Criterion30-piece study125-piece study
Best used forSetup validation, early development, low-risk runsProduction sign-off, high-risk or high-volume parts
Statistical confidenceModerate — a rough capability estimateHigh — reveals shifts, drift and distribution shape
Detects tool-wear drift?PoorlyYes, if data spans the wear cycle
Non-normal tailsUsually missedRevealed in the histogram
Cost / timeLowHigher
⚠️ 30 parts is noisy: a Monte Carlo study of 30-piece samples drawn from a process with a true Cp of 1.33 produced calculated Cp below 1.10 in about 5.5% of trials and below 1.20 in ~19.6%. Treat a 30-piece number as a rough estimate, and confirm with the larger study before customer sign-off. ISO 22514-3 (machine performance studies) explicitly warns against samples below ~30 observations and against running the study where tool wear is expected during collection.

Choosing measurement methods

💡 Gauge R&R first: AIAG-style guidance treats a GR&R of ≤10% of tolerance as acceptable and 10–30% as marginal. In deep holes the gauge is often harder than the part — an air gauge at the far end of a 1 m bore measures a different zone than a bore micrometer at the mouth. Fix the measurement system before you chart the process.

SPC by Characteristic

Bore diameter

Air gauge fast and non-contact; take high/low readings so roundness shows in the range. Diameter usually drifts down as the tool wears (negative skew) and jumps oversize when coolant pressure wavers (positive skew). Because the target diameter matters as much as the tolerance band, Cpm is the honest index here — it penalises a centred-but-off-target process.

Straightness & position

Straightness and true position are distances from a nominal line or point, so they follow a Rayleigh (or bivariate-normal) distribution — standard Cp/Cpk formulas do not apply directly. Use the Rayleigh or bivariate-normal methods for position capability. Straightness itself is best tracked with an I-MR chart; an off-square entry at the face is the single biggest cause of drift growing from a 0.1 mm start into 1+ mm at depth.

Surface roughness (Ra)

Ra data skews right and is driven by tool edge condition, cutting speed and vibration — not by tool-position error. Track it on an I-MR chart and watch for the trend rule, which reliably precedes visible surface degradation. Profilometer styli need regular verification; a worn stylus looks exactly like a worn tool on the chart.

Why deep hole data goes skewed

CauseEffect on distribution
Tool wear directionDiameter drifts down over tool life — negative skew
Coolant pressure fluctuationOccasional oversize holes at the cutting edge — positive skew
Material hard spotsRandom undersize bores — negative skew
Tool entry / pilot errorBell-mouth at entry pushes the upper tail — positive skew
⚠️ Skewed distributions: research on drilled-hole diameters finds a well-run process produces a near-normal distribution truncated on the left at the drill diameter and slightly right-skewed; as the process degrades, the mean, variance and skewness rise together while the lower percentiles barely move. Standard ±3σ limits on skewed data give false alarms on one side and missed signals on the other — use Cpk (one-sided), move to Cpm, transform the data, or work with a statistician on valid limits.

Out-of-Control Detection & Rules

Western Electric (1956) and Nelson (1984) run tests flag non-random patterns — the events that precede bad holes. The chart’s zones split the band into one-sigma widths: Zone C (within 1σ), Zone B (1–2σ), Zone A (2–3σ).

RuleTestSignal in a deep hole process
WE / Nelson 1One point beyond 3σGross shift — tool breakage, coolant loss, crash
WE 2 / Nelson 52 of 3 consecutive points beyond 2σ, same sideMean beginning to drift — early wear, coolant pressure drop
WE 3 / Nelson 64 of 5 consecutive points beyond 1σ, same sideBias developing — thermal drift, guide bushing wear
WE 4 / Nelson 28–9 consecutive points on one side of the centrelineProlonged bias — steady tool wear or a setup shift
Nelson 36 consecutive points increasing or decreasingTrend — the classic gundrill / BTA wear signature
Nelson 414 consecutive points alternating up and downTwo sources cycling — alternate machines, tools or operators
Nelson 715 consecutive points within 1σStratification — check sampling; pooled data from different states
Nelson 88 consecutive points all beyond 1σMixture — increased variation, two process levels combined
💡 Which rules to run: every rule you switch on adds false alarms. Rules 1–4 are recommended for routine use (combined false-signal chance below 1 in 100); add the 2-of-3 and 4-of-5 tests for earlier warning. For a tool-wear process the six-point trend (Nelson 3) is usually the first signal a gundrill or BTA head is past its life — plan a regrind when it appears. For autocorrelated data (a hole bore is not independent of the previous bore), rely on the point-beyond-limits rule, and consider EWMA or CUSUM charts to catch small sustained shifts early.

SPC Software & Machine Integration

Modern SPC software connects directly to CMMs, air gauges, digital calipers and machine PLCs, eliminating manual entry. Look for the following capabilities:

📊 Real-time chartingCharts update the moment a CMM, air gauge or caliper reads a bore — no data re-keying.
🔔 Alarm rulesWestern Electric / Nelson tests trigger alerts and a documented out-of-control action plan.
📈 Capability analysisAutomatic Cp/Cpk/Cpm/Ppk with histogram overlay and normality checks.
📑 PPAP reportingCapability reports formatted for customer submission and PPAP sign-off.
🎯 TraceabilityEvery measurement linked to the specific hole, tool serial, operator and shift.
🔧 Gauge integrationRS-232, USB, serial or OPC capture from CMMs, air gauges, calipers and PLCs.

Typical software stack

Adaptive control feeds SPC

Common SPC Mistakes in Deep Hole Drilling

MistakeConsequenceFix
Using specification limits as control limitsOperators tamper with stable points, or signals are missed entirelyControl limits from process data (mean ± 3σ), never from the drawing
Assuming in-control means in-specA stable but incapable process ships non-conforming boresEvaluate capability separately after establishing control
Targeting the mean instead of MMC/LMCRemoves material, adds cost, weakens the partAim the control target at MMC/LMC per the drawing, not the SPC mean
Forcing normality on tool-wear processesWrong limits, false alarms, missed wear signalsUse I-MR, transform the data, or apply the Clements percentile method
Single diameter reading per partRoundness error and wear stay hiddenRecord high/low values and plot the range
Running limits from the data foreverStale charts after tools, material or coolant changeRecalculate after any significant process change
Charting before the gauge is verifiedYou chart gage error, not process behaviourGR&R and calibration first
Applying Cp/Cpk to true positionMeaningless index — position is Rayleigh, not normalUse Rayleigh / bivariate-normal position capability
Over-adjusting on every pointOperators become the main source of variationAct only on signals; react to trends, not noise
⚠️ The most egregious error: substituting specification limits for control limits — especially on an X-bar chart, where specs apply to individual parts while the chart plots subgroup averages. Limits set too tightly cause tampering and added variation; limits set too loosely hide process change. The control limit is not the tolerance.

Hydraulic Valve Body Bore — 30-Piece Capability Study

Part:Hydraulic valve body, cast iron
Feature:Ø25 H7 bore — LSL 25.000, USL 25.021, target T 25.0105 mm
Method:BTA on a dedicated machine with coolant temperature control
Sample:30 consecutive parts, air-gauged; GR&R verified ≤ 10% of tolerance
Results:x̄ = 25.0115 mm, σ = 0.002 mm

Calculations

IndexCalculationValue
Cp(25.021 − 25.000) / (6 × 0.002) = 0.021 / 0.0121.75
Cpkmin[(25.021 − 25.0115), (25.0115 − 25.000)] / (3 × 0.002) = min[0.0095, 0.0115] / 0.0061.58
Cpm0.021 / (6 × √(0.002² + 0.001²)) = 0.021 / (6 × 0.002236) = 0.021 / 0.01341.57
✅ Reading the numbers: Cpk 1.58 lands inside the “capable” band (1.33–1.67) with an expected defect rate well under 10 ppm two-sided — acceptable for standard production SPC. Because Cpk < Cp (1.58 vs 1.75), the bore runs 0.001 mm high of target; Cpm catches that penalty and drops to 1.57. Now run the sensitivity test: if tool wear and coolant fluctuation push σ from 0.002 to 0.003, Cp falls to 1.17 and Cpk to 1.06 — marginal, ~1,000–2,000 ppm, and 100% inspection territory. A 0.001 mm change in sigma is the difference between world-class and inspection hell.
💡 Proven in practice: a cast-iron hydraulic valve body bore on a dedicated BTA machine with coolant temperature control and a strict regrind schedule ran Cp 3.29 — the process spread occupied only ~30% of the tolerance band. Automotive connecting-rod wrist-pin and crank bores on forged C70 steel with automated gaging ran Cp 2.14–2.57. Both are exactly what a disciplined SPC plus machine-capability program produces.

Keep Reading