Introduction
An operator measures a shaft and finds 25.03 mm against a target of 25.00 mm. He adjusts the machine down by 0.03 mm. The next part comes out at 24.96 mm, so he adjusts up by 0.04 mm. The part after that is 25.05 mm.
By the end of the shift, he has adjusted the machine eleven times, and the variation is worse than if he had touched nothing at all.
This is one of the most expensive mistakes in manufacturing, and it has a name. It is called tampering, and it happens when someone reacts to normal random variation as though it were a problem.
Statistical process control exists to answer one question with evidence instead of instinct: should I act on this reading, or leave the process alone?
For mechanical, production and industrial engineering students, SPC is one of the highest value topics in the quality syllabus. It appears in exams, in Six Sigma projects, in customer audits and in nearly every quality engineer job description you will ever read.
What Is Statistical Process Control (SPC)?
Statistical process control is a method of monitoring and controlling a manufacturing process using statistical techniques, mainly control charts, so that problems are detected while the process is running rather than after defective parts have been produced.
The key word is process, not product.
Traditional inspection looks at parts after they are made and sorts good from bad. SPC looks at the behaviour of the process over time and asks whether it is stable and predictable. A stable process produces good parts naturally. An unstable one will eventually produce bad ones no matter how much you inspect.
Think of it as the difference between checking a patient’s temperature once and watching the trend on a chart across a week. The single reading tells you where things stand now. The trend tells you what is about to happen.
Simple definition for your exam: Statistical process control is the use of statistical methods, particularly control charts, to monitor process variation over time, distinguish natural variation from unusual variation, and take corrective action before defective products are produced.
Origin and History of SPC
SPC was invented by Walter A. Shewhart at Bell Telephone Laboratories in the 1920s. In 1924 he produced the first control chart, and in 1931 he published the book that established the statistical basis for controlling manufactured product quality.
Shewhart’s key insight was the separation of variation into two types, which remains the foundation of the entire method.
W. Edwards Deming, who worked with Shewhart, carried the method into wartime American production and then into Japan in the 1950s, where it became a central part of the Japanese quality movement and later of the Toyota Production System.
SPC spread back into Western industry from the 1980s onward, and today it is a formal requirement in automotive quality standards, where suppliers must demonstrate statistical control and process capability on critical characteristics.
Common Cause vs Special Cause Variation
This is the concept the entire subject rests on. Get this right and everything else follows.
Common cause variation is the natural, random variation that is always present in a stable process. It comes from many small sources acting together, such as slight differences in material hardness, minor temperature changes, normal machine play and small measurement differences. It is built into the process, and it is predictable within a range.
Special cause variation is unusual variation caused by something specific that is not normally part of the process, such as a worn tool, a wrong setting, a new material batch, a machine fault or an untrained operator. It is unpredictable and it signals that something has changed.
The correct response is completely different for each.
| Type | Source | Predictability | Correct response |
|---|---|---|---|
| Common cause | Many small factors built into the process | Predictable within limits | Leave the process alone, improve the system if the range is too wide |
| Special cause | One specific identifiable event | Unpredictable | Investigate immediately and remove the cause |
Two classic errors come from confusing them.
Tampering means reacting to common cause variation as though it were special. The operator in the introduction was tampering, and his adjustments made things worse.
Ignoring a signal means treating special cause variation as normal noise. The tool was already worn, the chart showed a drift, and nobody investigated until the customer complained.
A control chart exists to tell these two apart with evidence rather than opinion.
What Is a Control Chart and How Does It Work?
A control chart is a time ordered plot of process data with three horizontal lines drawn on it.
Centre line (CL) is the average of the process. Upper control limit (UCL) is drawn three standard deviations above the average. Lower control limit (LCL) is drawn three standard deviations below the average.
Samples are taken at regular intervals, plotted in the order they were produced, and interpreted against these limits.
If all points fall randomly within the limits, the process is said to be in statistical control. It is stable and predictable, and it should be left alone.
If a point falls outside the limits, or the points form a non random pattern, a special cause is present and must be investigated.
Why three sigma? Because in a normal distribution, about 99.73 percent of natural variation falls within three standard deviations of the mean. So a point outside those limits is very unlikely to be natural variation, and far more likely to indicate a real change. Shewhart chose three sigma as a practical economic balance between reacting to false alarms and missing genuine problems.
Control Limits vs Specification Limits
This is the single most common mistake students make in exams and in interviews, so learn it clearly.
| Parameter | Control limits | Specification limits |
|---|---|---|
| Where they come from | Calculated from the actual process data | Set by the design drawing and customer requirement |
| What they describe | What the process is doing | What the customer will accept |
| Can they be changed by the engineer? | Only by changing the process | Only by a design change |
| What they tell you | Whether the process is stable | Whether the part is acceptable |
The two are completely independent. A process can be perfectly in control and still produce out of specification parts, if the control limits sit outside the tolerance band. Equally, a process can produce acceptable parts while being statistically unstable, which means it will fail sooner or later.
Never draw specification limits on a control chart. It encourages operators to adjust based on tolerance rather than process behaviour, which is exactly the tampering the chart was built to prevent.
Types of Control Charts in SPC
Control charts split into two families depending on the type of data you have. This classification is a very common exam question.
Variable Control Charts (for measured data)
Used when the characteristic can be measured on a continuous scale, such as diameter, length, weight, temperature or hardness.
X bar and R chart. The most widely used pair. The X bar chart tracks the average of each sample, and the R chart tracks the range within each sample. Used for sample sizes of roughly 2 to 9. The X bar chart shows whether the process centre is shifting, and the R chart shows whether the spread is changing.
X bar and S chart. Same purpose, but uses standard deviation instead of range. Preferred for larger sample sizes, usually above 9, because standard deviation uses all the data rather than just the highest and lowest values.
Individual and Moving Range chart, written as I-MR. Used when only one measurement is available at a time, such as in slow processes, batch chemical production or destructive testing.
Attribute Control Charts (for counted data)
Used when the characteristic is counted rather than measured, such as number of defects or number of defective units.
p chart for the proportion of defective units, with sample size allowed to vary. np chart for the number of defective units, with a constant sample size. c chart for the number of defects per unit, with a constant inspection area or unit size. u chart for the number of defects per unit, where the unit size varies.
Quick Selection Guide
| Data type | What you are tracking | Chart to use |
|---|---|---|
| Measured, sample size 2 to 9 | Average and spread | X bar and R |
| Measured, sample size above 9 | Average and spread | X bar and S |
| Measured, one reading at a time | Individual values | I-MR |
| Counted, proportion defective, varying sample | Fraction defective | p chart |
| Counted, number defective, constant sample | Count of defectives | np chart |
| Counted, defects per unit, constant unit | Defect count | c chart |
| Counted, defects per unit, varying unit | Defects per unit | u chart |
An easy way to remember the split: if you measure it, use a variable chart. If you count it, use an attribute chart.
How to Read a Control Chart: The Western Electric Rules
A point outside the control limits is the obvious signal, but it is not the only one. Non random patterns inside the limits also indicate a special cause. The standard set of detection rules comes from the Western Electric Statistical Quality Control Handbook.
Rule 1. One point falls outside the three sigma control limits. This is the classic out of control signal.
Rule 2. Two out of three consecutive points fall beyond two sigma on the same side of the centre line. Suggests a shift is beginning.
Rule 3. Four out of five consecutive points fall beyond one sigma on the same side. Suggests a sustained shift.
Rule 4. Nine consecutive points fall on the same side of the centre line. A clear run, indicating the process average has moved.
Rule 5. Six consecutive points steadily increasing or decreasing. A trend, often caused by tool wear, gradual heating or fatigue.
Rule 6. Fourteen consecutive points alternating up and down. Usually indicates two sources mixed together, such as two machines, two spindles or two operators.
Rule 7. Fifteen consecutive points within one sigma of the centre line. This looks good but is suspicious. It often means the control limits were calculated wrongly, or that data is being edited.
Rule 8. Eight consecutive points on both sides of the centre line with none within one sigma. Again suggests two different distributions mixed in one chart.
You do not need to memorise all eight for most exams, but rules 1, 4 and 5 come up constantly, and the interpretation of a trend as tool wear is a favourite interview question.
Process Capability: Cp, Cpk, Pp and Ppk
Control charts tell you whether the process is stable. Capability indices tell you whether the process is good enough for the tolerance it has to meet. Both are needed, and capability is only meaningful once the process is stable.
Cp compares the width of the specification band to the spread of the process. It answers whether the process is narrow enough to fit inside the tolerance. Its weakness is that it completely ignores where the process is centred.
Cpk measures the same thing but accounts for centring, by using the distance from the process average to the nearer specification limit. A process can therefore have a high Cp and a poor Cpk if it is shifted to one side.
Pp and Ppk are the long term performance versions of the same indices, calculated over a longer period using overall variation rather than within subgroup variation.
Typical interpretation:
| Cpk value | Meaning |
|---|---|
| Below 1.00 | Process is producing defects, not acceptable |
| 1.00 to 1.32 | Marginal, needs improvement |
| 1.33 | Common minimum requirement in the automotive industry |
| 1.67 and above | Good, often required for safety critical characteristics |
| 2.00 | Corresponds to Six Sigma level capability |
The relationship to remember: Cp is potential capability, Cpk is actual capability. If Cp is high and Cpk is low, the process spread is fine and the process simply needs centring, which is usually a quick fix.
Steps to Implement SPC in a Manufacturing Plant
Step 1: Select the characteristic to control. Choose critical to quality characteristics, usually taken from the drawing, the control plan or the process FMEA. You cannot chart everything, so chart what matters.
Step 2: Validate the measurement system first. Run a Gauge R and R study. If the measurement system itself is unreliable, every chart you draw afterwards is meaningless. This step is skipped constantly and it is the most common cause of failed SPC programmes.
Step 3: Choose the right chart. Variable or attribute, based on your data type and sample size, using the selection table above.
Step 4: Decide the sampling plan. Define subgroup size and sampling frequency. Subgroups should be rational, meaning parts within a subgroup are produced under conditions as similar as possible, so that variation between subgroups shows up clearly.
Step 5: Collect the initial data. Typically 20 to 25 subgroups before calculating limits, so the limits reflect the process rather than a lucky hour.
Step 6: Calculate the control limits. Compute the centre line and the three sigma limits from the collected data using the standard formulas and constants.
Step 7: Plot and study the chart. If special causes appear in the initial data, investigate them, remove the causes, and recalculate the limits from clean data.
Step 8: Assess capability. Once the process is stable, calculate Cp and Cpk against the specification.
Step 9: Train the operators. The person plotting the chart must understand what a signal means and what to do about it. A chart nobody can interpret is decoration.
Step 10: Define reaction rules and run the chart live. Write down exactly what action is taken when a rule is violated, who is informed and what gets recorded. Then review and recalculate limits whenever the process genuinely changes.
Benefits and Limitations of SPC
Benefits
- Detects problems while the process is running, before defective parts accumulate.
- Prevents tampering, which by itself often reduces variation immediately.
- Reduces the amount of inspection required, since a stable process needs less checking.
- Provides objective data for root cause analysis and improvement projects.
- Supplies the evidence customers and auditors ask for on critical characteristics.
- Builds process understanding, because the chart shows how the process actually behaves over time.
- Lowers scrap, rework and warranty cost.
Limitations
- Depends entirely on measurement quality. A poor gauge produces confident nonsense.
- Requires operator training and continuous discipline to keep charts filled and reviewed.
- Not well suited to very short production runs, where there is not enough data to establish limits.
- Charts often become a paperwork ritual, filled in without anybody acting on the signals.
- Attribute charts need large sample sizes to detect small changes.
- SPC monitors and detects. It does not by itself find the cause or fix the process.
- Wrongly calculated or never updated limits give false confidence.
SPC vs SQC vs Inspection
Students frequently confuse these three terms.
| Parameter | Inspection | Statistical Quality Control (SQC) | Statistical Process Control (SPC) |
|---|---|---|---|
| Focus | Individual parts | Lots and processes | The process over time |
| Timing | After production | During and after | During production |
| Main aim | Sort good from bad | Accept or reject lots and control quality | Keep the process stable and predictable |
| Typical tool | Gauges and instruments | Acceptance sampling plans | Control charts |
| Nature | Detection | Mixed | Prevention |
The clean relationship: SQC is the broad field of statistical quality methods, SPC is the part of it that monitors the running process, and inspection is what SPC aims to reduce the need for.
Where SPC Fits in Six Sigma, Lean and TQM
SPC is not a competing system. It is a shared tool used across all the major quality frameworks.
In Six Sigma, SPC belongs mainly to the Control phase of DMAIC, where control charts lock in the improvement so the gain does not fade. Capability indices are also used in the Measure phase to establish the baseline.
In lean manufacturing, SPC supports standard work and quality at the source, because a stable process is a prerequisite for just in time flow with low inventory.
In TQM, SPC delivers the fact based decision making principle, replacing opinion with data.
In ISO 9001 and IATF 16949, statistical techniques and demonstrated process capability on critical characteristics are formal requirements, and auditors specifically check whether charts are being acted upon rather than merely filed.
In root cause analysis, the control chart is often what triggers the investigation in the first place, by showing exactly when the process changed.
Frequently Asked Questions (FAQs) on Statistical Process Control
1. What is statistical process control in simple words?
It is a method of monitoring a process while it runs, using control charts, to see whether it is behaving normally.
It helps you decide whether to act on a reading or leave the process alone.
2. Who invented SPC and control charts?
Walter A. Shewhart at Bell Telephone Laboratories, who created the first control chart in 1924.
W. Edwards Deming later spread the method widely, especially in Japan.
3. What is the difference between common cause and special cause variation?
Common cause variation is the natural random variation always present in a stable process.
Special cause variation comes from a specific identifiable event such as a worn tool or a wrong setting.
Common cause needs system improvement, special cause needs immediate investigation.
4. What is the difference between control limits and specification limits?
Control limits are calculated from the actual process data and describe what the process is doing.
Specification limits come from the design drawing and describe what the customer will accept.
They are independent, and specification limits should never be drawn on a control chart.
5. Why are control limits set at three sigma?
Because about 99.73 percent of natural variation falls within three standard deviations of the average.
A point beyond that is very unlikely to be natural variation, so it is treated as a signal.
6. What are the main types of control charts?
Variable charts for measured data, including X bar and R, X bar and S, and I-MR.
Attribute charts for counted data, including p, np, c and u charts.
7. When should I use an X bar and R chart?
When you are measuring a continuous characteristic such as diameter or weight, with subgroup sizes of roughly two to nine.
8. What is the difference between Cp and Cpk?
Cp compares the process spread with the tolerance width but ignores centring.
Cpk accounts for both spread and centring, so it reflects actual capability.
9. What Cpk value is acceptable?
A Cpk of 1.33 is the common minimum requirement in the automotive industry.
Safety critical characteristics often require 1.67 or higher.
10. What is tampering in SPC?
It is adjusting a process in response to normal random variation.
Tampering almost always increases variation and is one of the main problems SPC was designed to prevent.
Conclusion
Statistical process control gives an engineer something rare on a shop floor: a defensible answer to whether a process needs attention right now.
For your exams, hold three anchors. The difference between common cause and special cause variation, and the correct response to each. The difference between control limits and specification limits, which is the most commonly examined point in this topic. And the chart selection logic, where measured data goes to variable charts and counted data goes to attribute charts.
For your interviews, be ready to interpret a chart out loud. Describe what a single point beyond the limit means, what a run of nine points on one side means, and why a steady upward trend usually points to tool wear. Then mention that you would validate the gauge before trusting any of it. That last point separates candidates who have used SPC from those who have only read about it.
For your career, remember what the method is really teaching. Most processes do not need constant adjustment. They need someone patient enough to watch the data, quiet enough to leave normal variation alone, and alert enough to act the moment something genuinely changes.