Reference card — verified IS codes & SPC constants
Verified 2026-09-14 against ~/Standards library + BIS portal. ConcreteInfo reference card.
Codes referenced
Frequency of sampling, sample size (3 cubes), 7d/28d testing, acceptance criteria.
Sampling, identity of samples, test results records.
Mean of 4 consecutive ≥ fck + 0.825σ; every individual result ≥ fck − 3.
Cube casting, curing, compression testing procedure.
Standard 3-sigma control limits; control chart factors A2, D3, D4.
Why statistical QC for concrete — variation is real
Concrete strength varies. Even at a well-controlled plant with identical materials and procedure, three cubes from the same load will rarely be identical. Sources of variation:
- Materials — aggregate gradation shift, cement strength lot variation, admixture sensitivity to temperature.
- Production — moisture in aggregate, weighing tolerance, mixing time, mixer wear.
- Sampling & testing — cube mould filling, compaction, curing temperature, testing machine alignment.
- Placing — transportation time, ambient temperature, addition of water at site.
A single cube result tells you almost nothing about the true strength of the concrete in place. IS 456 therefore requires sampling in groups of 3 cubes and acceptance based on the average. Statistical QC formalises this — X̄/R charts turn random variation into a visible, actionable signal.
IS 456 Cl 15.2 — sampling plan for strength
| Concrete grade | Sampling frequency | Sample size |
|---|---|---|
| M25 or lower | 1 sample per 10 m³ or part thereof; or 1 per 5 batches (whichever is more frequent) | 3 cubes · test 1 at 7d, 2 at 28d |
| M30 or higher | 1 sample per 5 m³ or part thereof; or 1 per 5 batches (whichever is more frequent) | 3 cubes · test 1 at 7d, 2 at 28d |
| Very large pours (raft, pile cap) | Add 1 sample per 50 m³ of additional QC | 3 cubes |
| Number of test results for σ (Cl 9.2.4.2) | Not less than 30 for mix-design target strength calculation | All data points |
X̄ chart — control limits at ±3σ
The X̄ chart plots the average of each subgroup (3 cubes for concrete). Control limits use the average range R̄:
UCL(X̄) = X̄̄ + A2 × R̄ · LCL(X̄) = X̄̄ − A2 × R̄
A2 is a tabulated control-chart factor based on subgroup size n. For n = 3 (the IS 456 default), A2 = 1.023.
| n (subgroup) | A2 (X̄ limits) | D3 (R lower) | D4 (R upper) | d2 (σ from R̄) |
|---|---|---|---|---|
| 2 | 1.880 | 0 | 3.267 | 1.128 |
| 3 (IS 456) | 1.023 | 0 | 2.575 | 1.693 |
| 4 | 0.729 | 0 | 2.282 | 2.059 |
| 5 | 0.577 | 0 | 2.114 | 2.326 |
| 6 | 0.483 | 0 | 2.004 | 2.534 |
R chart — within-batch variability
The R chart plots the range (max - min) within each subgroup. UCL(R) = D4 × R̄; LCL(R) = D3 × R̄. For n = 3: D4 = 2.575, D3 = 0. An R chart point above UCL indicates a special cause making one batch's 3 cubes more spread than usual — early signal of mix, sampling, or testing problem.
Acceptance criteria (complementary to charts)
Control charts track process stability; IS 456 Cl 16.1 acceptance criteria determine whether each batch's strength conforms to specification:
- Every individual cube result ≥ fck − 3 (Table 11)
- Mean of 4 consecutive samples ≥ fck + 0.825σ, or fck + 3 for M15 and fck + 4 for M20 and above, whichever is greater (Table 11)
- σ estimated from at least 30 prior test results (Cl 9.2.4.2); update as more data available
Source: IS 456:2000 §15.2 (frequency of sampling), §16.1 (acceptance criteria), §9.2.4.2 (σ from 30+ results).
8 Nelson out-of-control rules
A point beyond ±3σ is the basic signal (Rule 1). Nelson (1984) listed seven additional pattern-sensitive rules that catch process shifts earlier:
Basic signal — process mean or variation has shifted.
Strong early signal of mean shift.
Pattern of drift — process mean moving slowly.
Sustained mean shift — material or process changed.
Trend — gradual drift, often material stockpile or equipment.
Stratification — subgroups not capturing real variation.
Mixture — two different processes producing data.
Overdispersion — within-group variation much larger than expected.
Worked example — M30 site, 20 samples × 3 cubes
| Parameter | Value |
|---|---|
| Concrete grade | M30 (fck = 30 MPa, target mean 38.25 = 30 + 1.65 × 5) |
| Subgroups | 20 samples × 3 cubes each = 60 cubes (28-day) |
| Assumed σ from 30 prior results | 5.0 MPa |
| Overall mean X̄̄ (of 20 subgroup means) | 38.2 MPa |
| Average range R̄ (of 20 subgroups) | 4.1 MPa |
| A2 (n=3) | 1.023 |
| UCL(X̄) = X̄̄ + A2 × R̄ | 38.2 + 1.023 × 4.1 = 42.4 MPa |
| CL(X̄) | 38.2 MPa |
| LCL(X̄) = X̄̄ − A2 × R̄ | 38.2 − 1.023 × 4.1 = 34.0 MPa |
| D4 (n=3) | 2.575 |
| UCL(R) = D4 × R̄ | 2.575 × 4.1 = 10.6 MPa |
| LCL(R) | 0 MPa (D3=0 for n=3) |
| IS 456 Cl 16.1 acceptance | Individual ≥ 30 + 1.65×5 = 38.25 MPa · mean of 4 ≥ 30 + 0.82×5 = 34.1 MPa |
Interpretation: 2 sample means above UCL(X̄) = 42.4 MPa flag an investigation (Rule 1). The X̄̄ is well above target mean — mix is performing strongly. All R values below UCL(R) = 10.6 MPa — within-batch spread is in control. Recalculate σ from these 60 results; if σ is now lower (e.g. 4.2 MPa), mix can be optimised or margin reduced.
Common deviations seen on Indian sites
Engineer takes 1 cube per truck, declares strength on that 1 number. IS 456 requires 3 cubes per sample and the 4-sample mean criterion — a single high result does not justify release of the next 4 batches.
Site maintains only an X̄ chart (or worse, just cubes without chart). The R chart catches within-batch spread — a sudden rise means sampling, compaction, or testing problems even when mean is stable.
Two consecutive samples above UCL flagged but no action taken. Out-of-control signal means a special cause is present; ignoring it guarantees the next batch is also out of control.
Chart exists but the project spec doesn't say "investigate after Rule 1, halt pour after Rule 1+Rule 4 together". Without defined action rules, the chart is decoration, not a control tool.
Field-checklist blocks
A short questionnaire, an inspector's checklist, and three worked examples that show how the chart logic above resolves into a decision on the pour.
Questionnaire — ask before the first subgroup of the pour is accepted
- Is the sample a set of three cubes from one batch, so the subgroup size is 3 and the tabulated factors (A2 = 1.023, D4 = 2.575, D3 = 0) actually apply to what you are plotting?
- Has σ been established from at least 30 prior 28-day test results under IS 456 Cl 9.2.4.2, or are we still running on an assumed value that nobody has revisited?
- Is the sampling frequency for this pour taken from the table above, and are the samples spread across the pour rather than drawn from one convenient truck?
- Are both charts being maintained, X̄ for batch-to-batch drift and R for within-batch spread, or is there only a mean being recorded somewhere?
- Do the control limits get recalculated when the mix, the cement brand or the aggregate source changes, or are they still the limits from the first lot of subgroups?
- Is there a written action rule — who investigates a Rule 1 signal, and does the pour wait for that investigation?
- Are 7-day results kept out of the 28-day subgroup means, so the data feeding the chart is the data the acceptance criteria are judged on?
Inspector's checklist — what to verify on site
- Cube register: sample ID, pour location, grade, batch or ticket number, and cast date for every subgroup of three.
- Subgroup sheet: the three individual 28-day cube strengths, the subgroup mean and the range, with the X̄ and R point entered on the chart on the day the result lands.
- σ derivation note: where the σ in use came from, how many prior results it rests on, and the date it was last reviewed.
- Limit recalculation sheet: X̄̄, R̄, A2 = 1.023, D4 = 2.575 and D3 = 0, with the resulting UCL and LCL values written down, not just plotted.
- Nelson check per point: a written note of which of the 8 rules were tested against the new point and whether it was inside or outside the limits.
- Out-of-control record: cause, action taken, and the disposition of the concrete that the signal applies to.
- Action-rule document: what halts the pour, who may release it, and what happens to the strength already placed.
- Chart kept with the register it was built from: limits are only meaningful against the data that produced them.
What happens if… three worked examples
What happens. X̄̄ = 38.2 MPa and R̄ = 4.1 MPa give UCL(X̄) = 38.2 + 1.023 × 4.1 = 42.4 MPa, so the two subgroup means above 42.4 MPa are Rule 1 signals. Every one of the 60 cubes still clears the IS 456 Table 11 individual floor of fck − 3 = 27 MPa, and the group-of-4 mean gate of 34.0 MPa (fck + 0.825σ = 30 + 0.825 × 5 = 34.125, rounded off to the nearest 0.5, against fck + 4 = 34 MPa for M20 and above) is met comfortably. A chart signal and an acceptance failure are different things, and this is the pair that gets confused most often on site.
Likely outcome. The mix is strong; the process is not stable. You investigate the two batches (cement lot, aggregate moisture, admixture dosing) and you keep the existing limits until the cause is named. Recalculating limits around an unexplained high point hides the signal rather than closing it. If the recalculated σ from these 60 results comes in lower, around 4.2 MPa as the worked example above suggests, you then have room to optimise the mix or trim the margin.
Preventive correction. Tag the two batches with the delivery tickets, pull the admixture and moisture records for those dates, and re-plot the chart with the investigation note beside the points.
What happens. UCL(R) = 2.575 × 4.1 = 10.6 MPa, so that subgroup's range is an R-chart signal while its mean plots comfortably inside 34.0 MPa to 42.4 MPa. On a mean-only chart nothing shows at all. The cause is not the mix. It is sampling, compaction or testing: a badly filled mould, uneven tamping, a curing tank running out of temperature control, or platen alignment on the testing machine.
Likely outcome. A measurement-integrity signal, not a strength signal. The suspect cubes go back to the lab and the failure pattern is examined. A conical break is the qualifying mode; anything else is a flag. The testing machine gets checked before the mix does, because a reading you cannot trust cannot condemn a batch. Meanwhile the acceptance gates on the remaining cubes are applied as normal.
Preventive correction. Keep the R chart. It is the only part of this system that tells you the difference between concrete that is weak and testing that is wrong.
What happens. The limits on the chart were computed from the material in use before the change. The new cement moves the process mean, so the old limits now flag the new material as out of control, and the chart shows a step that has nothing to do with the concrete. With fewer than 25 subgroups the limits are provisional in any case — 25 subgroups, 75 cube results, is the stated minimum for stable limits on X̄.
Likely outcome. Two populations on one chart, and Nelson Rule 7 — eight points beyond 1σ on both sides — is what eventually exposes it. The honest treatment is a second chart on the new material. Recompute X̄̄ and R̄ from the change point, mark the change date on the old chart, and run the new one until it has its own 25 subgroups. Tighten the mean in the meantime. A chart that has quietly absorbed two materials is decoration.
Preventive correction. Put the change point on the chart the day the material changes, and treat the first 25 subgroups after it as a fresh series rather than as a continuation.
Frequently Asked Questions
What is the minimum sample size for an X̄ control chart on concrete strength?
How often should control limits be recalculated?
What is the difference between an X̄ chart and an individuals chart?
Related articles
Amit Haridas
Founder & Proprietor, ConcreteInfo. 25+ years experience in construction QA/QC, concrete technology, RMC plant operations, consulting and technical training across India. For X̄/R chart setup, mix-design σ validation, or statistical-QC programme design, contact amit@concreteinfo.in.
Technical updates, not sales emails
One email when a new article goes out on concrete technology, QA/QC and construction quality. No promotions, and your address is never sold or passed to advertising networks. Subscribing is separate from making an enquiry.