Lean Six Sigma · Interactive lesson
Short-Run SPC: Control Charts for High-Mix, Low-Volume Processes
Learn how to monitor one common process while products, grades, gauges, targets, and normal variation keep changing—an intermediate Lean Six Sigma application of statistical process control. You will transform mixed-product data, qualify valid product families, and select the correct short-run chart without mistaking planned product changes for process instability.
Why this matters at work: a conventional chart of raw values can signal every time the mill changes product. Separate charts can be equally weak when each product appears only a few times. Short-run SPC creates one statistically meaningful process view by removing the changes that were supposed to happen.
By the end, you should be able to
- Explain why short-run charts are named for short product runs—not for a chart that is used briefly.
- Recognize when a traditional control chart becomes misleading in a high-mix process.
- Calculate deviation-from-target and standardized short-run statistics.
- Select among X-nominal, DNOM, Z/W, standardized p, np, c, and u charts.
- Qualify products before pooling them into one common process stream.
- Apply short-run SPC to gauge, temperature, weld profiles, mechanical properties, and defect data in a steel mill.
Before you start
You should understand the basic purpose of a control chart, the difference between variables and attributes data, and the meaning of a mean and standard deviation. The companion lesson Control Charts, Explained provides that foundation.
Why short-run charts exist
A conventional control chart works best when the plotted values come from one stable process condition. In a high-mix operation, however, a schedule may look like this:
Problem 1: one raw chart
A raw gauge chart jumps whenever target gauge changes. Those jumps are planned product changes, but the chart can interpret them as shifts, trends, or excessive moving ranges.
Problem 2: one chart per product
Separate charts may contain only two or three observations before the product disappears for weeks. There is not enough sequential information to estimate or maintain reliable product-specific limits.
The short-run solution
Translate each result into a common statistical language before charting it. You can subtract the product target, divide by the product standard deviation, or compare the observed count with its product-specific expected count. The transformed stream then answers:
Key terminology
| Term | Plain-language meaning | Practical interpretation |
|---|---|---|
| Short run | A small quantity of one product before the process changes to another product. | The chart may run for years; the individual product campaigns are short. |
| Process stream | The common equipment, method, measurement system, and causal mechanism being monitored. | Products may differ, but they must still be generated by a defensibly common process. |
| Product family | A group of products similar enough to share a short-run baseline or transformation. | Same mill alone is not enough; route, control mechanism, measurement, and variation matter. |
| Nominal or target | The intended value for a specific product. | Subtract it to remove planned differences in product center. |
| Deviation from nominal | Actual result minus that product's target. | Positive means above target; negative means below target. |
| Standardized value | Deviation expressed in units of that product's standard deviation. | A value of +2 is two product-specific standard deviations above expectation. |
| Homogeneity of variance | Products have sufficiently similar natural process variation. | Required for simple DNOM/X-nominal pooling; otherwise standardize or separate. |
| Historical parameter | A validated mean, target, standard deviation, or baseline rate estimated before monitoring. | It must come from stable, representative data—not a mixture already containing special causes. |
| Rational subgroup | Measurements grouped so within-subgroup variation represents short-term variation. | Examples include several thickness readings across one coil or repeated specimens under one condition. |
Interactive: raw values versus coded values
The same twelve coils are shown below. Toggle the display to see why mixed raw values can hide the process story.
Raw gauge changes mainly reflect intentional target changes. A conventional moving range would be inflated at product transitions.
The two core transformations
1. Difference from nominal
- Di: coded deviation for observation i, in the original measurement unit
- Xi: observed result
- Tj: target for product j
Use when product targets differ but natural variation is sufficiently comparable.
2. Standardization
- μj: validated expected mean or appropriate target for product j
- σj: validated within-process standard deviation for product j
- Zi: distance from expectation in standard-deviation units
Use when both center and natural variation differ across products.
Formula explorer
Change the measurement, target, and product standard deviation. The result updates without reloading the page.
Interactive decision flowchart: select the short-run SPC chart
This rebuild preserves the logic of the source flowchart. Select any terminal chart to open its purpose, conditions, data requirements, formula, worked example, steel-mill use, and warnings.
How to use it: follow the arrows like a true flowchart. Click any terminal chart box to jump to the detailed explanation below.
Guided chart-selection wizard
Start
What are you trying to monitor?
Select a terminal above or complete the wizard. The detailed chart panel will open here.
Interactive: can these products share one short-run chart?
Short-run SPC begins with engineering qualification, not a formula. Check the statements that are true for the proposed product family.
Interactive: screen for comparable variation
The source flowchart asks whether product standard deviations are equal. In practice, use stable product-specific estimates, engineering judgment, graphical comparison, and—where appropriate—a formal equal-variance analysis. This quick ratio is a screening diagnostic only, not a universal acceptance test.
Screening result: similar enough to investigate a difference-from-nominal chart, subject to stable data and engineering review.
A defensible short-run SPC implementation procedure
- Define the monitoring question. Decide whether you are monitoring one product or the health of a common process across many products.
- Preserve time order. Arrange observations by actual production or test sequence. Product order is not a substitute for time order.
- Define the process stream. Specify equipment, route, operating mechanism, measurement system, test location, and subgroup logic.
- Form candidate product families. Use engineering knowledge before statistical pooling. Document exclusions and known process breaks.
- Validate historical baselines. Estimate targets, expected means, standard deviations, ranges, or defect rates from stable and representative periods.
- Choose the coding method. Use deviation from nominal only when product variation is comparable; otherwise standardize.
- Select the chart family. Follow variables versus attributes, individuals versus subgroups, and defective units versus defect counts.
- Build Phase I limits. Use enough pooled stable subgroups to estimate the common chart behaviour, investigate signals, remove only verified special causes, and recalculate once.
- Freeze the baseline for Phase II. Monitor new data using the validated historical parameters. Do not let every new point continuously redefine normal.
- Maintain traceability. Keep product, grade, gauge, heat, slab, rolling practice, test location, recipe, and equipment status available for signal investigation even though the chart plots coded values.
Worked steel-mill example: final-gauge deviation
Problem
A rolling mill alternates among three products with targets of 8.20, 7.80, and 8.40 mm. Historical within-product standard deviations are comparable, so an X-nominal chart is proposed. Twelve coils are produced in time order.
| Coil | Product | Target, mm | Actual, mm | Deviation, mm | Moving range, mm |
|---|---|---|---|---|---|
| 1 | A | 8.20 | 8.25 | +0.05 | — |
| 2 | B | 7.80 | 7.84 | +0.04 | 0.01 |
| 3 | C | 8.40 | 8.35 | −0.05 | 0.09 |
| 4 | A | 8.20 | 8.23 | +0.03 | 0.08 |
| 5 | B | 7.80 | 7.77 | −0.03 | 0.06 |
| 6 | C | 8.40 | 8.42 | +0.02 | 0.05 |
| 7 | A | 8.20 | 8.21 | +0.01 | 0.01 |
| 8 | B | 7.80 | 7.78 | −0.02 | 0.03 |
| 9 | C | 8.40 | 8.51 | +0.11 | 0.13 |
| 10 | A | 8.20 | 8.33 | +0.13 | 0.02 |
| 11 | B | 7.80 | 7.92 | +0.12 | 0.01 |
| 12 | C | 8.40 | 8.54 | +0.14 | 0.02 |
Build the chart one step at a time
Calculation
Interpretation: coils 10 (+0.13 mm) and 12 (+0.14 mm) exceed the target-centered UCL. Coils 9–12 also show a sustained positive shift. Investigate gauge-control bias, roll wear, target-table changes, temperature-model drift, or measurement changes around coil 9.
Where short-run SPC fits in a steel mill
Final gauge
Strong first application
Chart actual gauge minus target gauge. Use X-nominal for individuals or DNOM for rational subgroups when gauge variation is comparable; standardize when controllability changes by gauge family.
Exit or finishing temperature
Difference or standardized
Chart actual minus recipe target. Separate or standardize rolling practices, slab thicknesses, and grades when their thermal response or sensor variation differs.
Coiling temperature
Difference or standardized
Detect broad overcooling or undercooling across grades while retaining product identifiers for nozzle, water-flow, season, and pyrometer investigations.
Width
Often X-nominal
Use actual width minus ordered target when trimming or rolling variation is comparable across width families and the same measurement system is used.
Rolling force, torque, power
Model-residual chart
Expected values depend on grade, width, reduction, temperature, pass, speed, and roll condition. Chart actual − predicted, then standardize the residual if residual variation changes.
ERW weld profiles
Ratio or model residual
Account for OD, wall thickness, setup, and commissioning state before pooling OD width, center width, ID width, heights, or ferrite-line width.
Tensile properties
Standardized residuals
Do not center YS on SMYS automatically: SMYS is a lower requirement, not the process mean. Model grade, gauge, D/t, end, orientation, test location, and cooling condition.
Charpy energy
Use narrowly defined families
Account for specimen size, test temperature, orientation, location, grade, gauge, rolling practice, and slab thickness. Avoid pooling base metal, HAZ, and weld metal merely because all are measured in joules.
Defects and holds
Standardized p or u
Use standardized p for proportions defective and standardized u for defects per length, area, tonnes, or other changing inspection opportunity when baseline rates vary by family.
Ŷ is the expected response from a validated engineering or statistical model. Chart the residual only after checking that residuals represent one stable error mechanism.
Interactive: control limits are not specification limits
Use the four scenarios to separate process stability from product conformance.
The process is stable and its natural control band fits inside the customer specifications.
Control limits
Calculated from stable process behaviour. They answer: has the process changed?
Specification limits
Set by customers, codes, or engineering requirements. They answer: is the unit acceptable?
Interpretation guide: when you see this, it means this
| What you see | What it suggests | What to investigate |
|---|---|---|
| Raw values jump at every product change, coded values remain stable | The raw chart was reacting to intended target changes. | Use the coded short-run chart; keep product labels for traceability. |
| Coded values shift above zero across several products | A common positive process bias, not a single product target issue. | Calibration, control-loop bias, recipe table, common sensor, roll or cooling condition. |
| Only one product repeatedly signals after standardization | Product baseline may be wrong or the family assumption may fail. | Target/mean/σ estimate, route, test location, product-specific mechanism. |
| Location chart stable; W or range chart signals | Average performance is centered, but short-term variation has increased. | Measurement instability, material inconsistency, control oscillation, equipment wear. |
| DNOM chart signals mainly for products with larger σ | Equal-variation assumption is likely false. | Standardize by product variation or split the family. |
| Standardized p/u value exceeds +3 | Observed defect level is unexpectedly high after accounting for baseline rate and sample opportunity. | Defect definition, inspection exposure, product mix, process change, inspection sensitivity. |
| Chart stable but coded center is not zero | The process may be consistently biased relative to targets. | Decide whether to remove the bias, revise a technically justified target, or use a process-centered Phase II chart while preserving target performance reporting. |
Common mistakes and exam traps
1. Pooling because equipment is shared
Same mill does not prove the same process stream. Route, mechanism, measurement, and variation must be comparable.
2. Calling specifications control limits
Specifications judge conformance; control limits detect change. They are calculated differently and answer different questions.
3. Using SMYS as a tensile process mean
A minimum requirement is not the expected stable mean. Centering there can create a biased and uninformative coded chart.
4. Standardizing unstable history
Dividing by a standard deviation does not repair special causes. It simply hides them on a different scale.
5. Estimating σ from two or three observations
Small estimates are highly uncertain. Use qualified history, hierarchical pooling, or delay live charting until the baseline is defensible.
6. Mixing measurement systems
Different gauges, test frames, specimen geometries, or locations can create artificial shifts and false product variation.
7. Creating artificial subgroups
A subgroup is not just several rows. Members must represent the same short-term process condition.
8. Sorting by product
SPC requires time order. Sorting all Product A observations together destroys the sequence in which special causes occurred.
9. Moving ranges across invalid breaks
A shutdown, equipment rebuild, measurement change, or new process era may require a new baseline rather than one continuous MR stream.
10. Over-normalizing destructive tests
Sparse Charpy or tensile results may support periodic surveillance, but they should not be treated like high-frequency online process measurements without acknowledging delay and sampling uncertainty.
Excel and Minitab implementation
Recommended worksheet structure
| Column | Purpose | Example |
|---|---|---|
| Date/time or sequence | Preserve production order | 2026-07-29 08:15 |
| Product family | Defines the qualified common stream | Steckel gauge family 1 |
| Product ID | Retrieves target and parameters | 8⅝ × .323 |
| Actual result | Observed measurement or count | 8.25 mm |
| Target / expected mean | Product-specific location baseline | 8.20 mm |
| Product σ or rate | Product-specific variation or attribute baseline | 0.035 mm |
| Coded statistic | Deviation, Z, or standardized attribute residual | +1.43 |
| Traceability columns | Signal investigation | Heat, slab, rolling practice, gauge, end, tester |
Excel formulas
| Task | Example Excel formula | Notes |
|---|---|---|
| Look up product target | =XLOOKUP([@Product],Parameters[Product],Parameters[Target]) | Keep parameters in a controlled table. |
| Deviation from target | =[@Actual]-[@Target] | Preserves original unit. |
| Standardized variable value | =LET(x,[@Actual],mu,[@Mean],s,[@SD],IF(s>0,(x-mu)/s,NA())) | Guard against zero or missing σ. |
| Moving range | =ABS(CurrentDeviation-PreviousDeviation) | Do not bridge invalid process breaks. |
| Average by product | =AVERAGEIFS(Data[Actual],Data[Product],[@Product]) | Use only validated stable baseline rows. |
| Product standard deviation | =STDEV.S(FILTER(Data[Actual],Data[Product]=[@Product])) | Review sample size and stability. |
| Standardized p | =LET(p,[@Defectives]/[@n],p0,[@BaselineRate],(p-p0)/SQRT(p0*(1-p0)/[@n])) | Requires 0<p0<1 and a valid sample size. |
| Standardized u | =LET(u,[@Defects]/[@Opportunity],u0,[@BaselineRate],(u-u0)/SQRT(u0/[@Opportunity])) | Opportunity must use the same unit as u0. |
Minitab: standardized individual variables
- Arrange the individual measurements in one column and the product/part identifier in a second column.
- Choose Stat → Control Charts → Variables Charts for Individuals → Z-MR.
- Enter the measurement column under Variables and the product identifier under Part indicator.
- When validated historical means and standard deviations exist, open Z-MR Options → Parameters and supply them by part/product.
- Review both panels: the Z chart for location and the MR chart for changes in standardized short-term variation.
For X-nominal or other custom transformations, calculate the coded statistic in the worksheet first, then chart the coded column with the appropriate ordinary I-MR or X̄-R procedure using historical parameters where applicable.
Practice activity: select and calculate a standardized u-chart point
Scenario
A pipe mill monitors surface indications per 1,000 m. Product family P has a stable expected rate of 0.60 indications per 1,000 m. The current run covers 2,500 m and contains 6 indications.
- Choose the chart family.
- Calculate the observed defects per 1,000 m.
- Calculate the standardized statistic.
- Decide whether the point exceeds the +3 control limit.
Because 3.67 exceeds the +3 limit, the run is a special-cause signal after accounting for both the product baseline and the inspected length.
Check your understanding
Answer all five questions, then select Check my answers.
Final summary
- “Short run” refers to the limited production quantity of each product—not the operating life of the chart.
- Use a traditional chart when one repeatable product/process condition supplies enough sequential data.
- Use X-nominal or DNOM when targets differ but product variation is reasonably comparable.
- Use Z/W or standardized attribute charts when product-specific variation or baseline rates differ.
- Variables versus attributes, individual versus subgroup, defective versus defect, and constant versus varying opportunity determine the terminal chart.
- Qualify the process stream and product family before pooling. Common units or shared equipment are not sufficient.
- Control limits detect changes in the coded process; specifications still judge product acceptance.
- Keep complete product and process traceability so every coded signal can be investigated in engineering terms.
Suggested next lesson: Control Charts, Explained for signal rules, followed by a process-capability lesson once stability is established.
References and further reading
- McShane-Vaughn, M. E., The ASQ Certified Six Sigma Black Belt Handbook, short-run control charts decision flowchart and formula summary.
- Montgomery, D. C., Introduction to Statistical Quality Control, control-chart theory, rational subgrouping, and Phase I/Phase II practice.
- ISO 7870-8, Control charts — Part 8: Charting techniques for short runs and small mixed batches.
- NIST/SEMATECH, e-Handbook of Statistical Methods: Control Charts.
- Minitab Support, Overview for Z-MR Chart and Z-MR methods and formulas.
Lesson complete
You now have a structured Lean Six Sigma method for deciding whether to use a traditional chart, a difference-from-nominal chart, or a standardized short-run chart—and for defending the product family behind that decision.