Statistical Quality Control and Quality Analytics

Introduction and context

Construction quality decisions are based on evidence: concrete cube strengths, asphalt densities, aggregate gradations, soil moisture contents, dimensional measurements, weld results, and many other observations. A single test result describes one item, but quality control must judge a process and decide whether a lot, batch, or completed element is acceptable. Statistical quality control (SQC) provides a disciplined way to distinguish normal process variation from unusual conditions and to support decisions that are consistent, transparent, and technically defensible.

Construction measurements naturally vary. Some variation is caused by ordinary factors such as material properties, temperature, operator technique, equipment precision, and sampling location. Other variation arises from assignable causes, such as incorrect batching, poor curing, a damaged testing machine, inadequate compaction, or an unsuitable supplier. Statistical quality control does not remove variation; it helps the project team understand it and respond appropriately.

This chapter develops practical methods for diploma-level construction engineering: descriptive statistics, sampling, control charts, Pareto analysis, histograms, trend analysis, and basic process capability. These tools should complement—not replace—contract specifications, approved test methods, inspection plans, engineering judgement, and safety requirements.

Learning objectives

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

1. Explain common sources and types of variation in construction processes.
2. Calculate and interpret the mean, range, variance, and standard deviation of test results.
3. Distinguish a population from a sample and describe suitable sampling practices.
4. Construct and interpret basic control charts for measured quality characteristics.
5. Use histograms, Pareto charts, and trend analysis to identify priorities and unusual patterns.
6. Explain specification limits, control limits, and basic process capability indices.
7. Use statistical evidence responsibly when supporting acceptance or corrective-action decisions.

1. Variation, populations, and samples

1.1 Common and assignable causes

Common-cause variation is the inherent, combined effect of many small influences within a stable process. For example, concrete strengths from properly controlled batches will not be identical even when the mix design and procedures are unchanged. Assignable-cause variation is associated with a specific, identifiable change. Examples include a broken scale, a new untrained operator, excessive water added to concrete, or failure of a curing tank.

A process is statistically stable when its variation is predictable over time and no assignable cause is evident. Statistical stability does not necessarily mean that the process meets the specification. A stable process may consistently produce results that are too low. Conversely, a process may have an acceptable average but be unstable and therefore unreliable.

1.2 Population and sample

The population is the complete set of units or results of interest—for example, every concrete batch placed in a bridge deck. A sample is a subset selected for measurement. Since testing every unit may be destructive, expensive, or impractical, construction quality decisions usually depend on samples.

A sample must represent the population and must be collected according to an approved plan. Convenience sampling—testing only easily accessible or visually good material—can produce serious bias. Important controls include the sampling location, time, batch identity, quantity, preservation, test method, and chain of custody.

Random sampling gives each eligible unit a known opportunity for selection. Systematic sampling selects units at a planned interval, such as every tenth truck, provided that the interval does not coincide with a hidden periodic pattern. Stratified sampling divides the population into meaningful groups, such as different shifts, suppliers, or placement zones, and samples each group. In construction, stratification is often valuable because a project may involve several crews, sources, or environmental conditions.

1.3 Measurement quality

Statistical analysis cannot correct poor measurements. Accuracy describes closeness to the accepted reference value, while precision describes the closeness of repeated measurements to one another. A test system can be precise but inaccurate if it is consistently biased. Calibration, competent personnel, standardised procedures, equipment maintenance, duplicate testing, and proficiency checks are therefore essential.

2. Descriptive statistics for construction data

Descriptive statistics condense a dataset while retaining information useful for engineering interpretation.

2.1 Mean and median

For observations x₁, x₂, …, xₙ, the arithmetic sample mean is:

Here x̄ is the sample mean, xᵢ is observation i, and n is the number of observations. The mean has the same units as the observations and is appropriate for many approximately symmetric datasets, but it can be influenced strongly by an extreme result.

The median is the middle value after observations are arranged in order. If n is odd, it is the value in position (n + 1) ÷ 2. If n is even, it is the average of the two middle values. The median is less sensitive to outliers and is useful for skewed data, such as repair durations or defect costs.

2.2 Range and standard deviation

The range is:

It is simple but depends only on two observations. The sample variance and sample standard deviation are:

The denominator n − 1 is used when estimating the population variance from a sample. The standard deviation s is expressed in the same units as x and describes the typical spread around the sample mean. These formulas assume quantitative observations that are comparable and measured on the same basis. They should not be applied blindly to data containing different units, different test ages, or mixed populations.

The coefficient of variation is sometimes useful for comparing relative variation:

It is meaningful when x̄ is positive and not close to zero. For example, a standard deviation of 3 MPa has a different practical meaning when the mean strength is 20 MPa than when it is 80 MPa.

Worked example 1: concrete strength results

Eight 28-day compressive strengths, in MPa, are 34, 37, 36, 35, 39, 33, 36, and 40.

The mean is:

The range is 40 − 33 = 7 MPa. The squared deviations from 36.25 sum to 40.5 MPa². Therefore:

The coefficient of variation is approximately (2.41 ÷ 36.25) × 100% = 6.65%. The results have a mean of 36.25 MPa and moderate spread, but acceptance still requires comparison with the relevant specification, sampling rule, and test validity. A mean alone is not an acceptance criterion unless the contract explicitly makes it so.

3. Visual analysis of quality data

3.1 Histograms

A histogram groups numerical observations into adjacent intervals called bins. It shows the distribution’s centre, spread, skewness, possible outliers, and multiple peaks. A bimodal histogram may indicate two sources, crews, materials, or operating conditions that should not have been combined.

Bin widths should be selected so that meaningful patterns are visible without creating excessive noise. The analyst should record the units, sample size, period, and specification limits. A histogram is descriptive; it does not prove that a process is stable or that a product conforms.

3.2 Pareto analysis

A Pareto chart orders categories from the largest contributor to the smallest and often adds a cumulative-percentage line. It is useful for prioritising action, not for claiming that a fixed universal “80/20” relationship always exists. Categories must be defined consistently. For defects, the measure may be number of defects, rework hours, cost, or safety impact. These measures can produce different priorities.

For example, if observed defects are: honeycombing 18, low cover 10, misplaced sleeves 7, and surface cracks 5, the total is 40. Honeycombing represents 45%, and the cumulative share after low cover is 70%. The project team may first investigate concrete placement and vibration, while also considering whether a less frequent defect has a much greater safety consequence.

3.3 Trend and run analysis

A run chart plots results in time or sequence order. It can reveal drift, cycles, sudden shifts, or changes after an intervention. A steadily declining compaction density may indicate increasing moisture variation or deteriorating equipment. Alternating high and low results may suggest two crews, two test locations, or a measurement problem.

Trend analysis must respect the order of observations. Sorting values before plotting destroys time-related information. The analyst should annotate changes such as supplier substitution, weather events, equipment maintenance, mix adjustment, or a new operator.

4. Control charts

4.1 Purpose and interpretation

A control chart displays a quality statistic against time or production order. It normally includes a centre line and statistically derived upper and lower control limits. The centre line commonly represents the process mean. Control limits describe expected process variation when the process is stable; they are not the same as contractual specification limits.

A point outside a control limit is evidence that an assignable cause may be present. Additional warning patterns can include a sustained run on one side of the centre line, a clear trend, or unusual cycles. Rules should be defined before reviewing the data to reduce subjective interpretation. Any signal should trigger investigation, not automatic rejection of all material.

4.2 Individuals chart for individual measurements

When one measurement is obtained per time period, an individuals–moving range approach may be used. For consecutive observations, the moving range is:

The average moving range is:

For the common moving range of two observations, an estimate of within-process standard deviation is approximately:

The approximate three-sigma limits for the individuals chart are:

These relationships assume a reasonably stable process and roughly independent, approximately normally distributed measurements. The lower limit may be below a physically possible value; it is then interpreted as a statistical boundary, not as a negative physical result.

For subgrouped measurements, such as five concrete cylinders sampled from a batch, x̄–R or x̄–s charts may be used. Their constants depend on subgroup size and should be taken from a recognised statistical reference or validated software. Control limits must be calculated from appropriate baseline data and revised only under a documented procedure.

4.3 Control limits versus specification limits

Control limits are calculated from process behaviour. Specification limits are imposed by design, contract, code, or client requirements. A process can be in statistical control but fail the specification, or be outside statistical control while individual results happen to lie within specification. Acceptance decisions must use the contractual acceptance rule; control charts support process management and investigation.

5. Basic process capability

Process capability compares the natural spread of a stable process with specification limits. It should be used only after the measurement system is adequate and the process is reasonably stable. Capability indices are not a substitute for acceptance testing.

For a two-sided specification with lower specification limit LSL and upper specification limit USL, the potential capability index is:

The centred capability index is:

Here s is an appropriate estimate of within-process standard deviation. Cₚ measures the width of the specification interval relative to six standard deviations; it does not consider whether the mean is centred. Cₚₖ also considers the nearest specification limit. A common organisational benchmark is 1.33, but the required value must come from the project or organisation and must not be assumed universally.

For a one-sided upper limit, a corresponding index is:

For a one-sided lower limit:

Capability calculations assume that the observations represent a stable process and that the distributional model is reasonable. If data are strongly skewed, autocorrelated, mixed from different populations, or too few for a meaningful estimate, capability indices may mislead. Transformation, a distribution-free method, or separate analysis of homogeneous groups may be more appropriate.

Worked example 2: dimensional capability

Suppose a stable cutting process produces lengths with mean x̄ = 100.4 mm and sample standard deviation s = 0.20 mm. The specification is 100.0 to 101.2 mm.

The spread exactly equals the specification width, but the process is off-centre toward the upper side? More precisely, the mean is 0.4 mm above the lower limit and 0.8 mm below the upper limit, so the lower side is critical. The process requires centring and/or variation reduction. Capability does not itself authorise acceptance of any individual item.

6. Sampling and acceptance decisions

An acceptance plan specifies what is sampled, how often, which test method is used, the acceptance criterion, and what happens when a result is doubtful or nonconforming. The plan should be agreed before production wherever possible. Retesting should not be used selectively to replace an unfavourable valid result; the contract and test standard should govern confirmatory testing.

Sampling risk has two aspects. Producer’s risk is the risk that acceptable work is rejected. Consumer’s risk is the risk that unacceptable work is accepted. Increasing sample size generally improves information but increases cost and time. A sample average can hide an individual low result, while focusing only on the lowest result can ignore the overall process. The correct statistic depends on the specification.

Results should be traceable to the lot, location, time, sample collector, equipment, laboratory, and test method. Outliers must be investigated for transcription errors, specimen damage, invalid test conditions, or genuine process events. A result should not be discarded merely because it is inconvenient. If a test is invalid, the reason and authorised replacement procedure should be documented.

A practical decision sequence is:

1. Confirm that the sample and test are valid.
2. Identify the applicable lot and contractual criterion.
3. Check individual and summary results against the criterion.
4. Review control-chart, trend, and distributional evidence.
5. Investigate unusual results and determine affected material.
6. Obtain engineering review where safety, durability, or structural performance may be affected.
7. Record acceptance, conditional acceptance, rejection, repair, additional testing, or other authorised disposition.

7. Integrated illustrative case: concrete placement for a retaining wall

The following is a fictional illustrative case, not a project specification.

Background

A contractor places concrete for a retaining wall specified at 30 MPa characteristic 28-day strength. Eight batches are tested, producing strengths of 34, 37, 36, 35, 39, 33, 36, and 40 MPa. During the next placement, the site quality engineer records lower early results and observes that one crew has begun adding water at the point of discharge.

Problem

The project team must determine whether the original results indicate a stable process and how to respond to the observed change. It must not declare acceptance solely because the original average exceeds 30 MPa.

Analysis

The original sample mean is 36.25 MPa and standard deviation approximately 2.41 MPa. The range is 7 MPa. A histogram of the original results would show a compact group with no obvious extreme value, but eight observations are too few to establish long-term capability confidently. A run chart of future results is needed because the water addition represents a potential assignable cause.

The engineer verifies batch tickets, sampling records, cylinder identification, curing conditions, and laboratory calibration. The crew is interviewed, and the location of affected batches is mapped. The contract’s acceptance rule is reviewed, including requirements for low individual results and possible core testing.

Approach and solution

The site manager prohibits unapproved water addition, retrains the crew, checks slump-control procedures, and verifies admixture dosing. Subsequent results are plotted in production order. A control-chart signal in the post-change data is treated as evidence for investigation, not as an automatic statistical rejection rule. If affected concrete has already been placed, the engineer evaluates its identity, structural significance, and any required additional testing under the specification.

The quality team also uses a Pareto summary of observed defects. If most rework hours arise from poor vibration rather than strength, corrective action must address placement practice as well as batching. The histogram is updated by crew and supplier to detect whether combined data conceal two different processes.

Outcome

The process is accepted only after the authorised acceptance criteria are met and traceability is complete. The control-chart review shows whether the corrective action restored predictable behaviour. Lessons are incorporated into the inspection and test plan, toolbox briefing, and supplier review.

Learning points

The case demonstrates that descriptive statistics summarise evidence, while control charts examine stability. Specification limits govern contractual conformity. A process change must be investigated even if earlier averages were satisfactory. Reliable sampling, measurement assurance, and documented disposition are as important as calculation.

8. Technology, governance, and limitations

Spreadsheet software, laboratory information-management systems, mobile inspection applications, and dashboard tools can calculate statistics and display charts quickly. Automated alerts can help identify missing results, trends, or control-limit signals. However, software does not know whether data were sampled correctly, whether a test was valid, or whether two datasets belong to the same population. Formula cells, units, rounding, version control, access rights, and audit trails should be verified.

Artificial intelligence may assist with anomaly detection, image-based defect classification, or prioritisation of inspection resources. Its output remains decision support. Training-data bias, unexplained classifications, false alarms, cybersecurity, and accountability are significant concerns. Safety-critical acceptance decisions require competent human review and documented technical authority.

Common limitations include small samples, non-random sampling, autocorrelation, mixed populations, measurement bias, non-normal distributions, and misuse of control limits as specification limits. Statistical significance is not the same as engineering significance: a very small change may be statistically detectable but operationally unimportant, while a rare defect may be operationally critical despite limited data.

Best practice is to define the purpose of analysis, preserve data integrity, use validated methods, stratify when necessary, visualise data in time order, investigate causes rather than merely label results, and link every corrective action to a responsible person and verification date.

Chapter Summary

Statistical quality control converts construction test results into evidence for process understanding and quality decisions. The mean describes central tendency; the median resists extreme values; range and standard deviation describe variation; and the coefficient of variation expresses relative spread. Sampling must be representative, traceable, and supported by valid measurement systems.

Histograms reveal distributional form, Pareto analysis helps prioritise categories, and run charts expose trends and shifts. Control charts distinguish predictable process variation from possible assignable causes, but their control limits are not contractual specification limits. Basic capability indices, Cₚ and Cₚₖ, compare process spread and centring with specification limits and are meaningful only for stable, suitably measured processes. Acceptance decisions must follow the applicable contract and test standard, supported by engineering judgement, documentation, and ethical treatment of evidence.

Key Terms

Variation: Differences among measurements or units produced by a process.

Common cause: Ordinary source of variation inherent in a process.

Assignable cause: Specific, identifiable source of unusual variation.

Population: Complete set of units or observations of interest.

Sample: Subset selected from a population for measurement.

Mean: Arithmetic average of observations.

Median: Middle ordered observation, or average of two middle observations.

Range: Maximum observation minus minimum observation.

Standard deviation: Measure of typical dispersion around the mean.

Coefficient of variation: Standard deviation divided by mean, expressed as a percentage.

Histogram: Graph showing frequencies within numerical intervals.

Pareto chart: Ordered category chart used to prioritise contributors.

Run chart: Plot of observations in time or sequence order.

Control chart: Time-ordered chart with a centre line and control limits.

Control limit: Statistical boundary associated with expected process variation.

Specification limit: Contractual or technical boundary for conformity.

Process capability: Comparison of stable process performance with specification limits.

Cₚ: Potential capability index based on specification width and process spread.

Cₚₖ: Capability index considering both spread and process centring.

Traceability: Ability to link a result to its sample, method, equipment, and source.

Review Questions

Short Answer Questions

1. Distinguish common-cause variation from assignable-cause variation and give one construction example of each.
2. Why is representative sampling important in construction quality control?
3. What information can a histogram provide that a mean alone cannot?
4. Explain the difference between control limits and specification limits.
5. State why a capability index should not normally be calculated for an unstable process.

Long Answer Questions

1. Describe a complete procedure for analysing a set of concrete strength results from sampling through acceptance decision.
2. Explain the purposes and limitations of histograms, Pareto charts, and run charts in construction quality management.
3. Derive and interpret the sample variance and standard deviation formulas, including the reason for using n − 1.
4. Discuss how measurement-system quality, sampling design, and data traceability affect statistical conclusions.
5. Explain Cₚ and Cₚₖ, including their assumptions and how their interpretations differ.

Analytical / Application Questions

1. A stable process has x̄ = 52 MPa and s = 3 MPa, with limits of 45 and 60 MPa. Calculate Cₚ and Cₚₖ, and identify the principal improvement priority.
2. A run chart shows ten consecutive compaction results, with the last seven below the centre line. What possible causes would you investigate, and what evidence would you collect before deciding on corrective action?
3. A Pareto chart shows that a low-frequency defect causes most of the repair cost, while a high-frequency defect causes little cost. Explain how the project should prioritise action and why frequency alone is insufficient.

Multiple Choice Questions with Answers

Question 1

For the eight concrete strengths in the worked example, what is the sample mean?

A. 34.25 MPa

B. 36.25 MPa

C. 37.50 MPa

D. 40.00 MPa

Correct Answer: B. 36.25 MPa

Question 2

Which measure is generally least affected by one extreme observation?

A. Mean

B. Range

C. Median

D. Variance

Correct Answer: C. Median

Question 3

What is the main purpose of a control chart?

A. To replace the contract specification

B. To identify possible changes in process behaviour over time

C. To guarantee that every item conforms

D. To calculate construction quantities

Correct Answer: B. To identify possible changes in process behaviour over time

Question 4

A specification limit is best described as:

A. A statistically calculated process boundary

B. A contractual or technical conformity boundary

C. The average of all test results

D. The difference between the largest and smallest result

Correct Answer: B. A contractual or technical conformity boundary

Question 5

If the mean is 100.4 mm, standard deviation is 0.20 mm, and limits are 100.0 and 101.2 mm, what is Cₚₖ?

A. 0.67

B. 1.00

C. 1.33

D. 2.00

Correct Answer: A. 0.67

Question 6

Which chart orders defect categories from the largest contributor to the smallest?

A. Histogram

B. Individuals chart

C. Pareto chart

D. Scatter plot

Correct Answer: C. Pareto chart

Question 7

For consecutive individual observations, the moving range is calculated as:

A. The sum of all observations

B. The absolute difference between consecutive observations

C. The mean divided by the sample size

D. The specification width divided by six

Correct Answer: B. The absolute difference between consecutive observations

Question 8

Which condition is most important before using basic process capability indices?

A. The process should be reasonably stable

B. All results must be identical

C. The sample must contain only one observation

D. Specification limits must be ignored

Correct Answer: A. The process should be reasonably stable

Question 9

Adding unapproved water to concrete at discharge is an example of a possible:

A. Common-cause variation only

B. Assignable cause

C. Median calculation

D. Specification limit

Correct Answer: B. Assignable cause

Question 10

Why should data often be stratified by crew, supplier, or shift?

A. To conceal poor results

B. To increase the specification limit

C. To reveal differences that combined data may hide

D. To eliminate the need for sampling

Correct Answer: C. To reveal differences that combined data may hide