Control charts for variables
In the case of the variables chart, control must be exerted on both central tendency
and variability. A quality control analyst must be concerned about whether there
has been a shift in values of the performance characteristic on average. In addition,
there will always be a concern about whether some change in process conditions
results in a decrease in precision (i.e., an increase in variability). Separate control
charts are essential for dealing with these two concepts. Central tendency is
controlled by the X̅-chart, where means of relatively small samples are plotted on
a control chart. Variability around the mean is controlled by the range in the
sample, or the sample standard deviation. In the case of attribute sampling, the
proportion defective from a sample is often the quantity plotted on the chart. In the
following section, we discuss the development of control charts for the variables
type of performance characteristic.
̅ and R charts
X
➢ Mean and variance establish a basis for every distribution.
➢ To control a process, it is necessary to bring both; the process average and
process variability, under control.
➢ If the process average is under control but the variability is not under control
the process will produce items with large variation and hence no limits can be
obtained for the range of the process.
➢ Similarly, if the process has constant variation but different averages, it is not
possible to obtain an estimate of the process average.
➢ Hence both the charts are to be plotted simultaneously.
𝑋 ~ 𝑁(𝜇, 𝜎 2 )
̅) = μ, V(X
∴ E(X ̅) = σ2 ⁄√n
where μ and σ2 are population parameters and are known as the standard values
for the process.
R-chart
The 3σ-limits for R-chart will be:
E(R) ∓ 3σR ; where σR is the standard deviation of R
But from the distribution of range, it is known that
E(R) = d2σ and σR = d3σ
where d2 and d3 are constants which depend only on the sample size n.
Case I: If the standards are given i.e. the value of σ is known.
Let k samples each of size n be drawn from a production process at regular
intervals.
Let xij: jth observation from the ith sample; i = 1, 2, …., k; j = 1, 2, …., n
Sample Number Sample Observations Sample Mean Sample Range
1 x11, x12, …, x1n ̅
X1 R1
2 x21, x22, …, x2n ̅
X2 R2
:
:
:
k xk1, xk2, …, xkn ̅
X𝑘 Rk
∑𝑛𝑗=1 𝑥𝑖𝑗
̅
X𝑖 = ; 𝑖 = 1, 2, … . , 𝑘
𝑛
R i = (max xij − min xij ) ; 𝑖 = 1, 2, … . , 𝑘
j 𝑗
∑ki=1 ̅
X𝑖 ∑ki=1 𝑅𝑖
̿
X= ; 𝑅̅ =
k k
The 3σ-limits for R-chart will be
C. L. = d2σ
U. C. L. = d2σ + 3d3σ = (d2 + 3d3)σ = D2σ
L. C. L. = d2σ - 3d3σ = (d2 - 3d3)σ = D1σ
Case II: If the standards are not given i.e. the value of σ is unknown, it can be
estimated from the sample.
̅
R
̅ = d2 σ
E(R) = R ∴ σ
̂=
d2
The 3σ-limits for R-chart will be
̅
C. L. = E(R) = 𝑅
U. C. L. = E(R) + 3σR
3d3 R̅ 3d3
= R̅+ = (1 + ̅ = D4 R
)R ̅
d2 d2
L. C. L. = E(R) - 3σR
3d3 R̅ 3d3
̅
= R− = (1 − ̅ = D3 R
)R ̅
d2 d2
The values of d2, d3, D1, D2, D3 and D4 are tabulated for different values of n
= 2, 3, …, 20
̅ − chart
X
̅) ∓ 3√V(X
The 3σ-limits are: E(X ̅)
Case I: If the standards are given i.e. the values of μ and σ2 are known.
The 3σ-limits for X̅-chart will be
̅) = μ
C. L. = E(X
̅) + 3√V(X
U. C. L. = E(X ̅)
= μ + 3σ/√n
̅) − 3√V(X
L. C. L. = E(X ̅)
= μ − 3σ/√n
Case II: If the standards are not given i.e. the values of μ and σ2 are unknown, it
can be estimated from the sample as:
R̅
̿
μ̂ = X ; σ
̂=
d2
̅-chart will be
The 3σ-limits for X
̅) = μ̂ = X
C. L. = E(X ̿
̅) + 3√V(X
U. C. L. = E(X ̅)
3R̅
= X̿+ ̿ + A2R
= X ̅
d2 √n
̅) − 3√V(X
L. C. L. = E(X ̅)
3R̅
̿
= X− = X̿ − A2R ̅
d2 √n
The values of A2 are tabulated for different values of n = 2, 3, …, 20
Procedure of X ̅ and R charts
✓ Compute 𝑅̅ for the given data and plot R-chart.
✓ If all the points are within the control limits for R-chart proceed to X ̅-chart.
✓ If the process shows lack of control i.e. if some points are outside the control
limits, compute the revised control limits for R-chart on the basis of those
sample points which are within the control limits.
✓ Continue this procedure till R-chart shows control i.e. the process is in control
with respect to its variability.
✓ Compute X ̿ using, ONLY THOSE SAMPLE POINTS WHICH ARE WITHIN THE
CONTROL LIMITS ON THE R-chart, and plot X ̅-chart.
✓ If the process shows lack of control i.e. if some points are outside the control
limits on the X ̅-chart, eliminate these points and repeat the entire procedure
from the beginning.
✓ If all the points on both the charts are within the control limits, then the process
is said to be in STATISTICAL CONTROL and the values are taken as standard
values for future use.
̅ and S charts
X
The range is efficient as an estimator for σ, but this efficiency decreases as the
sample size gets larger. Hence X̅ and S (standard deviation) charts are used when
either
1) The sample size n is moderately large, say n > 10
2) The sample size n is variable.
2
√∑ni=1(Xi − X)
where S =
n−1
𝑋 ~ 𝑁(𝜇, 𝜎 2 )
̅) = μ, V(X
∴ E(X ̅) = σ2 ⁄√n
where μ and σ2 are population parameters and are known as the standard values
for the process.
S-chart
S2 is an unbiased estimator for σ2 but S is not unbiased for σ, i.e. E(S) ≠ σ.
But from the distribution of standard deviation, it is known that
E(S) = c4σ and V(S) = (1-c42)σ2
where c4 is a constant which depends only on the sample size n.
Case I: If the standards are given i.e. the value of σ is known.
The 3σ-limits for S-chart will be
C. L. = c4σ
U. C. L. = c4σ + 3σ√1 − 𝑐42 = [𝑐4 + 3√1 − 𝑐42 ]σ = B6σ
L. C. L. = c4σ - 3σ√1 − 𝑐42 [𝑐4 − 3√1 − 𝑐42 ]σ = B5σ
Case II: If the standards are not given i.e. the value of σ is unknown, it can be
estimated from the sample.
Let k samples each of size n be drawn from a production process at regular
intervals.
Let xij: jth observation from the ith sample; i = 1, 2, …, k; j = 1, 2, …, n
Sample Sample
Sample Mean Sample Standard Deviation
Number Observations
1 x11, x12, …, x1n ̅
X1 S1
2 x21, x22, …, x2n ̅
X2 S2
:
:
:
k xk1, xk2, …, xkn ̅
X𝑘 Sk
∑𝑛𝑗=1 𝑥𝑖𝑗
̅
X𝑖 = ; 𝑖 = 1, 2, … , 𝑘
𝑛
2
∑𝑛𝑗=1(𝑥𝑖𝑗 − 𝑥𝑖 )
𝑆𝑖 = √ ; 𝑖 = 1, 2, … , 𝑘
𝑛−1
∑𝑘𝑖=1 𝑆𝑖
E(S) = S̅ =
𝑘
The statistic 𝑆⁄𝑐4 is an unbiased estimator of σ.
The 3σ-limits for S-chart will be
C. L. = E(S) = 𝑆̅
U. C. L. = E(S) + 3√V(S)
3S̅√1 − 𝑐42 3√1 − 𝑐42
= S̅ + = (1 + ) S̅ = B4 S̅
𝑐4 𝑐4
L. C. L. = E(S) - 3√V(S)
3S̅√1 − 𝑐42 3√1 − 𝑐42
= S̅ − = (1 − ) S̅ = B3 S̅
𝑐4 𝑐4
The values of 𝑐4 , B3 and B4 are tabulated for different values of n = 2, 3, …, 20
̅ − chart
X
̅) ∓ 3√V(X
The 3σ-limits are: E(X ̅)
Case I: If the standards are given i.e. the values of μ and σ2 are known.
The 3σ-limits for X̅-chart will be
̅) = μ
C. L. = E(X
̅) + 3√V(X
U. C. L. = E(X ̅) = μ + 3σ/√n
̅) − 3√V(X
L. C. L. = E(X ̅) = μ − 3σ/√n
Case II: If the standards are not given i.e. the values of μ and σ2 are unknown, it
can be estimated from the sample as:
S̅
̿
μ̂ = X ; σ
̂=
c4
̅-chart will be
The 3σ-limits for X
̅) = μ̂ = X
C. L. = E(X ̿
̅
̅) + 3√V(X
U. C. L. = E(X ̿ + 3S
̅) = X ̿ + A3 S̅
= X
c n 4√
3S̅
̅) − 3√V(X
L. C. L. = E(X ̅) = X
̿− ̿ − A3 S̅
= X
c4 √n
The values of c4 and A3 are tabulated for different values of n = 2, 3, …, 20
Procedure of X ̅ and S charts
✓ Compute 𝑆̅ for the given data and plot S-chart.
✓ If all the points are within the control limits for S-chart proceed to X ̅-chart.
✓ If the process shows lack of control i.e. if some points are outside the control
limits, compute the revised control limits for S-chart on the basis of those
sample points which are within the control limits.
✓ Continue this procedure till S-chart shows control i.e. the process is in control
with respect to its variability.
✓ Compute X ̿ using, ONLY THOSE SAMPLE POINTS WHICH ARE WITHIN THE
CONTROL LIMITS ON THE S-chart, and plot X ̅-chart.
✓ If the process shows lack of control i.e. if some points are outside the control
limits on the X ̅-chart, eliminate these points and repeat the entire procedure
from the beginning.
✓ If all the points on both the charts are within the control limits, then the process
is said to be in STATISTICAL CONTROL and the values are taken as standard
values for future use.
Natural Tolerance Limits and Specification Limits
If the process is in statistical control, then the control limits or the 3σ-limits (μ ∓
3σ, where μ is the process average and σ is the process standard deviation) are
called the NATURAL TOLERANCE LIMITS.
The width of these limits (6σ) is called NATURAL TOLERANCE of the process
and it indicates the variability of the process.
If μ and σ are not known then the Natural Tolerance Limits are given as 𝜇̂ ∓ 3𝜎̂
and 6𝜎̂ is the Natural Tolerance.
It may so happen that even though the process is shown to be in statistical control,
from the control charts, the customer may not be satisfied with the product. This
happens when the process does not meet the SPECIFICATION LIMITS as desired
by the customer. These specification limits are usually given in terms of upper and
lower specification limits.
A decision whether the customer will be satisfied with the product, whenever the
process shows statistical control, can be made by comparing the Natural Tolerance
Limits with the Specification Limits.
Let Xmax and Xmin denote the Upper Specification Limit (USL) and the Lower
Specification Limit (LSL) respectively; then the difference Xmax - Xmin is called
SPECIFIED TOLEREANCE.
Comparison between Natural Tolerance and Specification Tolerance can be made
by considering the following three cases:
Case I: Xmax - Xmin > 6σ i.e. Specified Tolerance is more than Natural Tolerance
Xmax (USL)
UCL
CL
LCL
Xmin (LSL)
In this case, as long as the process is in statistical control, the items produced
will also meet the required specifications. X ̅ can then be allowed to fluctuate
to the extent that the sample points are within the specification limits. This
will save time and money for frequent machine settings and delays due to
identifying for assignable causes of variation. If the ratio of (Xmax - Xmin) to
6σ is very large, it means that the process is too good for the product and
the specification limits can be brought closer to produce a product of
superior quality.
Case II: Xmax - Xmin = 6σ i.e. Specified Tolerance Limits coincide with Natural
Tolerance Limits
This is an ideal situation and, in this case, a process in statistical control
implies that the product meets the specifications. The process mean or
average is exactly the same as the specified mean and hence a slight
fluctuation in process average may result in some items of the product
falling outside the specification limits. As soon as such a SHIFT IS
DETECTED in the process average, immediate remedial action should be
taken.
Case III: Xmax - Xmin < 6σ i.e. Natural Tolerance is more than Specified Tolerance
UCL
Xmax (USL)
CL
Xmin (LSL)
LCL
In this case, even if the process is in statistical control, some defective items
not meeting the specifications may be produced. A slight shift in the process
average will further increase the number of defectives, hence it becomes
necessary to readjust the process with respect to process mean and process
variation or one may check whether it is possible to relax the Specified
Tolerance to the extent of Natural Tolerance. If 100% inspection is possible
the defective items can be sorted out. But if 100% inspection is not possible
the only alternative is relaxation of Specification Limits.
Change or Shift Detection
If the process average changes suddenly, the probability of a sample point falling
within 3σ limits will no longer be expected to be 0.9973 but it will depend on the
new process average.
Let Pa(p1) = P(a sample point falls within the control limits when the process
average p has been shifted to a new value p1)
= P(LCL ≤ p ≤ UCL)
= P(p ≤ UCL) - P(p ≤ LCL)
𝑈𝐶𝐿−𝑝1 𝐿𝐶𝐿−𝑝1
= 𝑃 (𝑍 ≤ 𝑝 𝑞
) − 𝑃 (𝑍 ≤ 𝑝 𝑞
)
√ 1𝑛 1 √ 1𝑛 1
E. g. 1) A manufacturing process is being monitored with the help of a p-chart
having Central Line at p̅ = 0.05 and n = 100. If the process average shifts from
0.05 to 0.08, what is the probability that the shift will be detected on the (i) first
sample, (ii) third sample, after the shift has occurred?
p̅ = 0.05, n = 100, p1 = 0.08, ∴ q̅ = 0.95
C. L. = p̅ = 0.05
̅̅̅̅
pq
U. C. L. = p̅ + 3√ n = 0.11538
̅̅̅̅
pq
L. C. L. = p̅ − 3√ =0
n
Let Q = P( not detecting a shift) = P(accepting a lot with mean = p1 = 0.08)
= Pa(0.08) = P(LCL ≤ p ≤ UCL)
𝑈𝐶𝐿−𝑝1 𝐿𝐶𝐿−𝑝1
= 𝑃 (𝑍 ≤ 𝑝 𝑞
) − 𝑃 (𝑍 ≤ 𝑝 𝑞
)
√ 1 1 √ 1 1
𝑛 𝑛
= P(Z ≤ 1.3) – P(Z ≤ -2.94) = 0.9016
Let P = P(detecting a shift) = 1 – Q = 0.0984
P(detecting a shift at the first sample) = P = 1 – Q = 0.0984
P(detecting a shift at the third sample)
= P(not detecting a shift at the first sample) P(not detecting a shift at the second
sample) P(detecting a shift at the third sample) = Q2P = 0.079987643
2) A process average shows control with R ̅ = 10 and X̿ = 60. If the process average
̿1 =65, find the probability of detecting a shift from X
shifts to X ̅-chart on the first
sample assuming no change in the value of R ̅. Given that n = 5, A2 = 0.577.
̅ = 10, X
R ̿ = 60, X
̿1 =65
̅-chart will be
The 3σ-limits for X
̿ = 60
C. L. = X
3𝑅̅
̿ + A2 R
U. C. L. = X ̅ = 65.77 𝜎̂ = 𝐴2
̿ − A2 R
L. C. L. = X ̅ =54.23
Let Q = P( not detecting a shift)
̿1 =65)
= P(accepting a lot with mean = X
̿ ≤ UCL)
= P(LCL ≤ X
̿− X
X ̿1 65.77−65 ̿− X
X ̿1 54.23−65
=𝑃( ̅
3𝑅
≤ 3×10 )− 𝑃( ̅
3𝑅
≤ 3×10 )
𝐴2 0.577 𝐴2 0.577
= P(Z ≤ 0.0148) – P(Z ≤ -0.2071) =
Let P = P(detecting a shift) = 1 – Q =
P(detecting a shift at the first sample) = P = 1 – Q =
XXXXX
