Statistical Process Control

Introduction and Context

Statistical process control (SPC) is a systematic method for monitoring, controlling, and improving processes through the statistical analysis of measured process data. In mechanical engineering, SPC is used in machining, casting, forming, heat treatment, assembly, inspection, and maintenance. Its purpose is not merely to identify defective products after manufacture; it is to understand process behaviour and prevent defects by responding to process changes at their source.

A process may produce parts that vary in dimension, surface finish, hardness, mass, or functional performance. Some variation is unavoidable. SPC distinguishes variation that is inherent to the process from variation caused by identifiable disturbances. This distinction supports rational decisions: continue production when the process is stable, investigate when an unusual cause appears, and improve the process when its natural variation is too large for customer requirements.

SPC is closely related to metrology and quality control. Reliable measurement systems are essential because a control chart can only be as trustworthy as the data used to construct it. Before applying SPC, the engineer should confirm instrument calibration, resolution, repeatability, reproducibility, and measurement uncertainty.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain common-cause and special-cause variation.
2. Establish rational subgroups for process monitoring.
3. Construct and interpret variable control charts for means, ranges, and standard deviations.
4. Construct and interpret attribute control charts for defective units and defects.
5. Distinguish control limits from specification limits.
6. Recognize non-random patterns and select appropriate process responses.
7. Evaluate process capability after statistical stability has been established.
8. Apply SPC principles in a manufacturing case.

1. Variation and Process Control

1.1 Common and special causes

Common-cause variation results from the cumulative effect of many small, continuously present influences. Examples include normal machine vibration, small differences in raw material, ordinary tool geometry variation, and routine environmental fluctuations. When only common causes are present, the process is statistically stable or predictable, although it may not meet specifications.

Special-cause variation arises from an identifiable and unusual source. Examples include a broken or incorrectly set tool, incorrect material grade, a coolant failure, a measuring instrument fault, or an operator changing a machine setting. Special causes make the process unpredictable until they are removed or incorporated deliberately into the process design.

The distinction is important because the wrong response can increase variation. Adjusting a stable process after every small fluctuation is called tampering. It often creates additional variation. Conversely, ignoring a special cause allows a correctable problem to continue.

1.2 Control versus capability

A process is in statistical control when its observed variation is consistent over time and no evidence indicates a special cause. A process is capable when its stable output can satisfy engineering specifications with an acceptable margin. A process can therefore be:

• stable and capable;
• stable but incapable;
• unstable but apparently within specifications at a particular moment; or
• unstable and incapable.

Control limits are calculated from process data and describe expected process behaviour. Specification limits are established by design, function, safety, customer requirements, or regulation. Control limits are not specification limits and must never be substituted for them.

2. Rational Subgroups and Data Structure

A rational subgroup is a set of observations collected under conditions that are as homogeneous as practical, so that variation within the subgroup primarily represents short-term common-cause variation. Differences between subgroup averages then reveal changes over time or between process conditions.

For example, if four consecutive shafts are measured every hour, the four measurements may form a subgroup. This arrangement is useful when the objective is to detect a shift in the process mean. Alternatively, one shaft from each of four machines may not form a rational subgroup if the purpose is to assess each machine separately, because machine-to-machine differences would be mixed with within-machine variation.

Subgrouping depends on the engineering question:

Sampling frequency should reflect the speed of the process and the likely time scale of special causes. Frequent sampling is justified when a fault can develop quickly or when the cost of producing nonconforming parts is high. Subgroup size commonly ranges from 3 to 5 for \bar{x}–R charts, although the appropriate value depends on the process and the statistical design.

Observations must be time-ordered. Randomly mixing data from different periods can conceal shifts, trends, and cycles. Subgroup definitions, sampling intervals, machine identity, operator, material lot, and measurement method should be recorded.

3. Variable Control Charts

Variable data are measured on a numerical scale, such as diameter, length, pressure, temperature, hardness, or mass. The most widely used charts for subgrouped variable data are:

• \bar{x} chart: monitors subgroup means;
• R chart: monitors subgroup ranges;
• s chart: monitors subgroup standard deviations.

The range chart or s chart is generally examined first because the estimate of process variation must be reliable before interpreting the chart for means.

3.1 Basic statistics

For subgroup i containing n observations x_i1, x_i2, ..., x_in, the subgroup mean is:

The grand mean for k subgroups is:

The subgroup range is:

The average range is:

The sample standard deviation of subgroup i is:

The average standard deviation is \bar{s} = \sum s_i/k. These formulas assume that observations within each subgroup are measured using a consistent method and represent comparable process conditions.

3.2 Mean and range charts

For a subgroup size n, the control limits for the mean and range charts are commonly written as:

A₂, D₃, and D₄ are control-chart constants determined by subgroup size. The lower range limit is often zero when D₃ is zero. The constants assume approximately independent observations within rational subgroups and are derived using the distribution of the sample range, commonly under a normality approximation.

For a subgroup size of 5, typical constants are A₂ = 0.577, D₃ = 0, and D₄ = 2.114. These values should be taken from an accepted statistical quality-control reference rather than recalculated casually.

3.3 Mean and standard-deviation charts

For larger subgroups, the \bar{x}–s chart is often preferred:

A₃, B₃, and B₄ depend on subgroup size. The standard-deviation chart generally uses more information than the range chart and is preferable when subgroup sizes are moderately large.

3.4 Worked example: shaft diameter

An engineer measures five consecutive shaft diameters in each of six initial subgroups. The subgroup means, in millimetres, are 20.02, 20.01, 19.99, 20.00, 20.03, and 20.01. The corresponding ranges are 0.04, 0.03, 0.05, 0.04, 0.06, and 0.03 mm.

The grand mean is:

The average range is:

For n = 5, use A₂ = 0.577, D₃ = 0, and D₄ = 2.114. Therefore:

The sixth subgroup mean, 20.01 mm, and range, 0.03 mm, are within their limits. This does not prove that the process is capable of meeting design requirements; it indicates only that the subgroup is consistent with the estimated process behaviour. The calculation also assumes that the initial subgroups are reasonably free of special causes. If an initial subgroup is affected by a known abnormal event, it should be investigated and treated according to a documented procedure rather than removed merely to improve the chart.

4. Attribute Control Charts

Attribute data classify observations or count events. They are useful when measurement is difficult, expensive, or naturally categorical.

4.1 p chart: fraction nonconforming

A p chart monitors the fraction of units that are nonconforming. If subgroup i contains nᵢ units and dᵢ are nonconforming:

The average fraction nonconforming is:

For constant or varying subgroup sizes, the limits are:

Limits below zero are set to zero. These limits use a binomial model: each unit is classified as conforming or nonconforming, units are appropriately comparable, and the probability of nonconformance is reasonably constant when the process is stable.

4.2 np chart: number nonconforming

An np chart monitors the number of nonconforming units when the subgroup size n is constant:

The np chart is easier for operators to interpret when every sample contains the same number of units.

4.3 c and u charts: defects

A nonconforming unit fails to meet a requirement, whereas a defect is a specific departure from a requirement. One unit may contain several defects.

A c chart monitors the number of defects per constant inspection unit:

A u chart monitors defects per inspection unit when the opportunity or area inspected varies:

The c and u charts are based approximately on a Poisson model, in which defects occur independently at a stable average rate. Overdispersion, clustering, or changing inspection opportunity can invalidate this assumption.

5. Interpreting Control Charts

A point outside a control limit is strong evidence of a special cause, but points inside limits do not automatically establish stability. Non-random patterns can also signal special causes.

Common supplementary rules include:

• one point beyond a 3σ control limit;
• two of three successive points beyond the same 2σ zone;
• four of five successive points beyond the same 1σ zone;
• eight successive points on one side of the centre line;
• six successive points steadily increasing or decreasing;
• a sustained cycle, stratification, or repeating pattern.

These rules increase sensitivity but also increase false alarms when many rules are applied. An organization should specify the rules before reviewing data and should consider the chart type, sampling frequency, and consequences of false alarms.

A practical investigation sequence is:

1. Confirm that the data and calculations are correct.
2. Check the measurement system and instrument status.
3. Identify the time, machine, operator, material lot, tool, and environmental conditions.
4. Contain potentially nonconforming product using a documented procedure.
5. Correct the special cause if verified.
6. Determine whether affected observations should remain in the baseline.
7. Restart or recalculate limits only under an approved rule.
8. Document the cause, action, verification, and disposition of product.

A point outside a limit is a signal for investigation, not automatic proof that every associated part is defective. Product acceptance must be determined using specifications and inspection requirements.

6. Process Capability

Capability indices should be calculated only after the process is stable and the measurement system is adequate. For a stable approximately normal process with upper specification limit USL and lower specification limit LSL:

where σ is the within-process standard deviation. Cₚ describes potential capability if the process is centred. The centred capability index is:

where μ is the process mean. Cpk reflects both spread and displacement from the specification centre. Higher values indicate greater margin, but the acceptable threshold depends on the organization, industry, safety risk, and customer requirements. Capability indices do not replace engineering judgement, distributional checks, or direct estimates of nonconformance.

For an R-chart estimate, a common estimate of within-subgroup standard deviation is:

where d₂ is a subgroup-size constant. This estimate is appropriate only under the assumptions associated with the range method and a stable process.

7. Implementation in a Manufacturing Organization

An effective SPC system has technical and organizational components. The technical component includes a defined characteristic, sampling plan, chart selection, calculation method, reaction plan, and record system. The organizational component includes trained personnel, authority to stop or contain production, management support, traceability, and periodic review.

A suitable implementation sequence is:

1. Select a critical or high-risk characteristic.
2. Define the measurement method and verify its adequacy.
3. Identify the process stream and rational subgroup.
4. Collect baseline data in time order.
5. Investigate obvious special causes before estimating limits.
6. Select a variable or attribute chart.
7. Calculate centre lines and control limits.
8. Publish a clear reaction plan at the workplace.
9. Review chart performance and capability.
10. Improve the process and update documentation when justified.

Digital SPC systems can automatically collect measurements, calculate limits, and issue alerts. However, automation does not eliminate the need for engineering judgement. Algorithms may generate false alarms because of autocorrelation, tool-change cycles, batch effects, missing data, or measurement drift. AI-based anomaly detection can supplement conventional charts, but its alerts should be explainable, validated, access-controlled, and connected to an accountable human response.

8. Illustrative Case Study: CNC Turning Line

Background

A production line machines a shaft diameter specified as 25.000 ± 0.025 mm. Five consecutive shafts are measured every hour. The line has previously experienced tool wear and occasional incorrect tool offsets.

Problem

The quality department reports intermittent nonconforming shafts. Final inspection detects the problem after several hours of production, causing sorting and rework.

Analysis

The engineering team verifies the micrometer using reference standards and conducts a short repeatability and reproducibility study. The measurement system is adequate for the decision. A \bar{x}–R chart is selected because the characteristic is continuous and five consecutive parts can be sampled economically.

The subgrouping plan is defined as five consecutive shafts from one machine, sampled every hour. Machine identity, tool number, material batch, operator, and offset are recorded. Initial data reveal a gradual downward trend in subgroup means while ranges remain relatively constant. The trend is consistent with tool wear rather than random common-cause variation.

Approach and solution

The reaction plan requires the operator to:

• stop and identify the last acceptable subgroup when a trend rule is triggered;
• verify the instrument and remeasure the subgroup;
• inspect tool condition and offset records;
• segregate product made since the last confirmed acceptable check;
• correct or replace the tool;
• take a confirmation subgroup before releasing production.

The process engineer also establishes a preventive tool-change interval based on measured wear, while retaining the chart to verify that the interval is effective. A separate chart is used after a major process change until a new stable baseline is approved.

Outcome

The number of parts requiring sorting decreases because the chart detects the developing condition before the diameter crosses the specification limit. The process mean is subsequently adjusted closer to the nominal value, and capability is reassessed using stable data. The improvement is not credited to the control chart alone; it results from the combination of a sound measurement system, rational subgrouping, a practical reaction plan, and corrective action addressing tool wear.

Learning points

The case illustrates that a chart is a decision-support system rather than a substitute for process knowledge. A stable range chart does not guarantee a stable mean. A trend may be more informative than a single point beyond a limit. Finally, the chart should be connected to product containment and corrective action, not treated as an isolated statistical display.

9. Limitations, Governance, and Best Practice

SPC has limitations. Non-normal data, autocorrelation, changing subgroup sizes, mixed process streams, low defect rates, and unstable measurement systems can produce misleading limits. A control chart cannot correct an incapable design, compensate for poor calibration, or identify a cause that has not been measured or recorded.

Control limits should not be recalculated every time a point is inconvenient. Such practice masks signals and makes the process appear stable. Limit changes should be justified, documented, and linked to a genuine process change or an approved baseline procedure.

Quality records should preserve chart history, revisions, investigation outcomes, and product-disposition decisions. Access to electronic SPC data should be controlled to prevent unauthorized alteration. In regulated or safety-critical industries, retention, traceability, electronic signatures, and validation requirements may apply.

Best practice includes:

• verify the measurement system before charting;
• define rational subgroups using process knowledge;
• keep data in chronological order;
• separate control limits from specification limits;
• use a documented reaction plan;
• investigate signals without blaming individuals prematurely;
• avoid tampering with stable processes;
• review chart performance and capability together;
• use stratification when multiple machines, tools, or materials are involved;
• validate automated alerts and maintain human accountability.

Chapter Summary

Statistical process control uses time-ordered data to distinguish predictable common-cause variation from unusual special-cause variation. Rational subgroups are designed so that within-subgroup variation represents short-term process behaviour and between-subgroup changes can reveal shifts. Variable charts include \bar{x}, R, and s charts; attribute charts include p, np, c, and u charts. Each chart requires appropriate data, assumptions, and control-limit calculations.

Control limits describe process behaviour, whereas specification limits describe product requirements. A process should be judged for capability only after its stability and measurement system have been established. Control-chart interpretation includes out-of-limit points and non-random patterns such as runs, trends, and cycles. Effective SPC requires a defined sampling plan, verified measurement, documented reaction procedures, product containment, root-cause investigation, and disciplined governance. Digital and AI tools can improve responsiveness but cannot replace engineering judgement or accountability.

Key Terms

Common cause: A routine source of variation inherent in a stable process.

Special cause: An identifiable, unusual source of process variation.

Statistical process control: Use of statistical methods to monitor and improve process behaviour.

Control chart: A time-ordered graph with a centre line and statistically derived control limits.

Control limit: A calculated boundary used to judge whether process behaviour is statistically unusual.

Specification limit: An engineering or customer requirement defining acceptable product performance.

Rational subgroup: A group of observations selected to represent comparable short-term process conditions.

Variable data: Quantitative measurements on a numerical scale.

Attribute data: Classification or count data.

Nonconforming unit: A unit that fails to meet one or more requirements.

Defect: A specific departure from a requirement; one unit may contain multiple defects.

\bar{x} chart: A chart for monitoring subgroup means.

Range chart: A chart for monitoring within-subgroup ranges.

p chart: A chart for monitoring the fraction nonconforming.

np chart: A chart for monitoring the number nonconforming at constant subgroup size.

c chart: A chart for monitoring defects per constant inspection unit.

u chart: A chart for monitoring defects per varying inspection opportunity.

Process capability: The ability of a stable process to meet specifications.

Cpk: A capability index reflecting both process spread and centring.

Tampering: Unnecessary adjustment of a stable process in response to random variation.

Review Questions

Short Answer Questions

1. Distinguish common-cause variation from special-cause variation.
2. Why must rational subgroups be defined before control limits are calculated?
3. What is the difference between a control limit and a specification limit?
4. When is an np chart preferred to a p chart?
5. Why should capability indices be calculated only after stability is assessed?

Long Answer Questions

1. Explain the construction and interpretation of an \bar{x}–R chart, including its assumptions and reaction procedure.
2. Compare p, np, c, and u charts, stating the type of data and sampling conditions appropriate for each.
3. Discuss how poor measurement-system performance can distort SPC conclusions.
4. Explain why tampering can increase process variation and describe how a reaction plan prevents it.
5. Describe an SPC implementation plan for a CNC machining process from characteristic selection through continuous improvement.

Analytical / Application Questions

1. A process produces subgroups of five measured parts. The grand mean is 50.00 mm and the average range is 0.10 mm. Using A₂ = 0.577, calculate the mean-chart limits and explain what a subgroup mean of 50.07 mm indicates.
2. In ten samples of 100 components, a total of 40 are nonconforming. Calculate \bar{p} and the approximate three-sigma p-chart limits for a sample size of 100. Explain why negative lower limits are replaced by zero.
3. A stable chart shows eight consecutive subgroup means below the centre line while all points remain within the limits. Identify the likely interpretation, list two possible engineering causes, and propose an investigation plan.

Multiple Choice Questions with Answers

Question 1

Which statement best describes common-cause variation?

A. Variation caused by a broken tool

B. Variation inherent in the routine process system

C. Variation caused by an incorrect material batch

D. Variation caused by a measuring instrument failure

Correct Answer: B. Variation inherent in the routine process system

Question 2

Which chart is designed to monitor subgroup means?

A. R chart

B. p chart

C. x̄ chart

D. c chart

Correct Answer: C. x̄ chart

Question 3

A p chart monitors:

A. The number of defects per unit

B. The fraction of units that are nonconforming

C. The range of measured values

D. The process standard deviation

Correct Answer: B. The fraction of units that are nonconforming

Question 4

For a constant subgroup size, which chart monitors the number of nonconforming units?

A. np chart

B. u chart

C. s chart

D. R chart

Correct Answer: A. np chart

Question 5

If a process has USL = 10.10, LSL = 9.90, and σ = 0.025, what is Cₚ?

A. 0.67

B. 1.00

C. 1.33

D. 2.67

Correct Answer: C. 1.33

Question 6

What should normally be examined first when using an x̄–R chart?

A. The specification limits

B. The range chart

C. The customer complaint rate

D. The process capability index

Correct Answer: B. The range chart

Question 7

Which situation is most appropriate for a u chart?

A. Constant sample size and number nonconforming

B. Variable inspection opportunity and defects per unit

C. Continuous measurements in subgroups of five

D. Constant sample size and number nonconforming units

Correct Answer: B. Variable inspection opportunity and defects per unit

Question 8

A point outside a control limit should initially be treated as:

A. Proof that every product is defective

B. A signal requiring investigation

C. A reason to delete the point

D. Evidence that specifications are incorrect

Correct Answer: B. A signal requiring investigation

Question 9

What is a rational subgroup?

A. A random mixture of all production data

B. A group selected under comparable short-term conditions

C. A subgroup containing only defective units

D. A group selected after viewing the chart pattern

Correct Answer: B. A group selected under comparable short-term conditions

Question 10

Why can frequent adjustment of a stable process be harmful?

A. It removes all common causes permanently

B. It causes tampering and may add variation

C. It makes specification limits wider

D. It converts variable data into attribute data

Correct Answer: B. It causes tampering and may add variation