Syntha Lab RU

Control chart: the defect share (p) and the means of measurements (X̄ and S) with rules and Cpk

From batches builds a control chart of the share of defects or returns (p) or a chart of the means and the spread of measurements (X̄ and S), calculates the center line and the 3σ limits, finds signals by four rules and from the mean charts calculates the capability indices Cp and Cpk against your tolerance. The limits can be recalculated without the points beyond the limits.

Choosing a factory and quality For whom: Quality specialists and buyers who watch defects and sizes of lots at the factory.

Calculate on your data →

At a glance

What question does it answer?
Whether the process is stable or a special cause has appeared and whether the process fits the tolerance.
What data is needed?
Batches in order: checked and defective or readings.
How to read the result?
Example on the default values (the share chart, 20 batches of 200): the average defect share 3.18%, the upper limit 6.89%; batch 13 (18 defective) is beyond the limit, signals 1. On the mean chart the same signal and Cpk 1.75. Example on the default values: Average share (center line): 3.18%; Upper limit at the average batch size: 6.89%; Lower limit at the average batch size: 0%; Batches: 20. The figures are illustrative: enter your own to see your result.

Why and how to reason

Defects always fluctuate, and it is important to tell ordinary noise from a real signal. A control chart does this. For the defect share by batch the center line is p̄ = total defects ÷ all items: here 3.18%. The 3σ limits are built as p̄ ± 3 × √(p̄ × (1 − p̄) ÷ n): for a batch of 200 units the upper limit is 6.9%, the lower zero. While batch shares are within the limits and behave randomly, the process is statistically in control, and there is no need to change it because of one bad batch. A point beyond the limit, eight points in a row on one side of the center, six in a row rising or falling, or two out of three far from the center are signals that something in the process has changed. For measurements (size, weight) there is a chart of means X̄ and spread S, and the indices Cp and Cpk show whether the process fits the tolerance.

When to use it

It is useful to monitor defects, returns by batch and measured product parameters at the factory or in receiving. Do not use it on a small number of batches (fewer than 20): the limits will be unstable. Assumptions: batches are homogeneous, measurements are independent, the process is stable in the period used to compute limits, the tolerance (LSL, USL) is set by product requirements, not by wish. The chart works well as an appendix to a factory contract: the parties agree on signal rules in advance.

How to read the result

First see whether there are signals: one point beyond the limit (out of 20 batches) and one batch with a signal under any rule. Find the cause in the log: a change of fabric lot, a new operator, a machine. After the cause is removed recompute the limits without that point (“recalculate” mode). If the process is in control but Cpk is below 1.33, it is “stably bad”: change the process itself rather than look for culprits. Cp shows the potential of the process, Cpk takes into account the offset of the center. If Cp is high and Cpk low, the process just needs to be shifted to the center of the tolerance.

Common mistakes

The usual mistakes: computing limits from bad batches, confusing control-chart limits with tolerance (they are different things), reacting to every point, not recording the cause of a signal, mixing batches of different products on one chart. If batch sizes differ, the limits change: do not use one number.

What data is needed?

  • Batches in order: checked and defective or readings.
  • The chart type and, for means, the tolerance on the size.
  • At least 20 batches as the base of the limits.

Inputs

FieldUnitValue in the exampleMeaning
Chart typeDefect or return share (p)Share: n items checked in a batch, d defective. Measurements: up to five readings per batch, for example width or length; the sample size must be the same.
LimitsFrom all pointsIf the series already had a special cause (a point beyond the limit) the limits counted from all points are inflated. The recalculation excludes such points once.
Lower tolerance23.9The lower tolerance limit on the measured size from the tech pack or the acceptance standard: for Cp and Cpk on the mean chart. Empty: the indices are not calculated.
Upper tolerance24.3The upper tolerance limit on the measured size. Empty: the indices are not calculated.

How it is calculated

StepFormula
Centrep̄ = Σ d ÷ Σ n
Limitsp̄ ± 3 × √(p̄ × (1 − p̄) ÷ n)
Rules1: beyond 3σ; 2: 8 in a row on one side; 3: 6 in a row rising or falling; 4: 2 of 3 beyond 2σ

Worked numbers

Default values from the app; in the interactive version you enter your own.

Average share (centre line)
3.18 %
Upper limit at the average batch size
6.89 %
Lower limit at the average batch size
0 %
Batches
20
Largest share in a batch
9 %
Points beyond the 3σ limits
1
Batches with a signal by any rule
1
The process is statistically in control
no: there are signals
The p chart: the share by batch and the limits
Line chart: Share: 3 → 2; Centre: 3.18 → 3.18; Upper limit: 6.89 → 6.89; Lower limit: 0 → 00246810Batch 1Batch 6Batch 11Batch 16Batch 20Share, Batch 1: 3Share, Batch 20: 2Centre, Batch 1: 3.18Centre, Batch 20: 3.18Upper limit, Batch 1: 6.89Upper limit, Batch 20: 6.89Lower limit, Batch 1: 0Lower limit, Batch 20: 0%Batches in order
  • Share
  • Centre
  • Upper limit
  • Lower limit
Data as a table
Batches in orderShareCentreUpper limitLower limit
Batch 133.186.890
Batch 22.53.186.890
Batch 34.53.186.890
Batch 423.186.890
Batch 53.53.186.890
Batch 62.53.186.890
Batch 733.186.890
Batch 843.186.890
Batch 923.186.890
Batch 102.53.186.890
Batch 113.53.186.890
Batch 1233.186.890
Batch 1393.186.890
Batch 142.53.186.890
Batch 1523.186.890
Batch 1633.186.890
Batch 173.53.186.890
Batch 182.53.186.890
Batch 1933.186.890
Batch 2023.186.890

Batches and signals

BatchShare, %Lower limitUpper limitRules
Batch 1306.89—
Batch 22.506.89—
Batch 34.506.89—
Batch 4206.89—
Batch 53.506.89—
Batch 62.506.89—
Batch 7306.89—
Batch 8406.89—
Batch 9206.89—
Batch 102.506.89—
Batch 113.506.89—
Batch 12306.89—
Batch 13906.891
Batch 142.506.89—
Batch 15206.89—
Batch 16306.89—
Batch 173.506.89—
Batch 182.506.89—
Batch 19306.89—
Batch 20206.89—

The 3σ limits show the natural spread of the process and not the tolerance: a process in control can produce defects and one out of control can be within tolerance. A signal means “look for a special cause”: a change of supplier, a fabric lot, a set-up. Rules: beyond 3σ; 8 in a row on one side; 6 in a row rising or falling; 2 of 3 beyond 2σ. The first 20–25 batches are better counted separately as the base of the limits. Footwear, accessories and children’s have their own defect share and tolerances.

Nuances and limits

  • Taking the 3σ limits for the tolerance: they are different things.
  • Counting limits on a series with known upsets: they are inflated, recalculate without them.
  • Reacting to every point: look for a cause only on a signal.

Terms used

Calculate on your data →