IOP Conference Series: Materials Science and Engineering

PAPER • OPEN ACCESS

Modified SPC for short run test and measurement process in multi-
stations
To cite this article: C K Koh et al 2018 IOP Conf. Ser.: Mater. Sci. Eng. 328 012009

View the article online for updates and enhancements.

This content was downloaded from IP address 158.46.212.8 on 16/03/2018 at 00:40

ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

Modified SPC for short run test and measurement process in

multi-stations

C K Koh1*, J F Chin1 and S Kamaruddin2

1
School of Mechanical Engineering, Universiti Sains Malaysia, 14300, Nibong Tebal,
Pulau Pinang, Malaysia
2
Department of Mechanical Engineering, Universiti Teknologi PETRONAS, 32610,
Seri Iskandar, Perak, Malaysia

*Email: [email protected]

Abstract. Due to short production runs and measurement error inherent in electronic test and
measurement (T&M) processes, continuous quality monitoring through real-time statistical
process control (SPC) is challenging. Industry practice allows the installation of guard band
using measurement uncertainty to reduce the width of acceptance limit, as an indirect way to
compensate the measurement errors. This paper presents a new SPC model combining
modified guard band and control charts (Z chart and W chart) for short runs in T&M process in
multi-stations. The proposed model standardizes the observed value with measurement target
(T) and rationed measurement uncertainty (U). S-factor (Sf) is introduced to the control limits
to improve the sensitivity in detecting small shifts. The model was embedded in automated
quality control system and verified with a case study in real industry.

1. Introduction
The key characteristics of test and measurement (T&M) manufacturing are short production runs of
multi-product families and testing at multi-stations. Classical Shewhart control charts, namely x̄ chart
and R chart have been widely used in statistical process control (SPC) to detect process variable shifts
in mean and variance [1]. Short production runs render these charts inefficacious as inherent meager
data do not warrant meaningful control limits [2]. Furthermore, Costa and Castagliola [3] underscore
the problem of growing risk of false acceptance in SPC due to measurement error. This leads to
consequences such as unnecessary process adjustment and loss of confidence in SPC.
In these premises, this research proposes a modified SPC model which combines modified guard
band and control charts (Z chart and W chart) to address the issues caused by short production runs
and measurement errors. The proposed model standardizes the observed value with measurement
target (T) and rationed measurement uncertainty (𝑈). The organization of the paper is as follow: A
review of the literature of measurement uncertainty and short run techniques and will be presented in
the next section. The proposed model and the industrial case study are described in section 3 and
section 4 respectively. Section 5 discusses the results obtained from the case study. Finally, section 6
draws conclusion.

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

2. Literature Review

2.1. Short Run approaches

Several short run charts have been proposed using statistics techniques such as difference chart,
DNOM (deviation from nominal) chart and standardized charts [1,4]. Difference chart (also called X-
nominal) registers differences between the mean to its nominal/target value [5]. DNOM chart is based
on ratios of subgroup means to nominal [6]. Useful in a scenario with shifting variances, standardized
charts with short runs techniques was proposed by Bothe [2] to revise the target value proportionally
to the amount of dispersion from product to product. It allows the operator to plot different product
families on a single x̄ chart and R chart.
Wheeler [5] extends the approach using the Z chart (also called Zed chart) and W chart to measure
central tendency and dispersion respectively. Consider a process where xij represents the ith response
on the jth part family member, i = 1, . . ., m, and j = 1, . . ., p. The process response is normalized using
an estimate of the population standard deviation, denoted by  ̂x𝑗 as:
𝑅𝑗

̂ x𝑗 =
𝑑2
(1)

where 𝑅𝑗 is the average range for the jth part family member; d2 is a statistic constant for subgroup
size (n) displaying a normally distributed quality characteristic. The Zij values for Z chart are then
calculated as:
𝑥𝑖𝑗 −𝑇𝑗
𝑍𝑖𝑗 = (2)

̂ x𝑗

where Tj is the target value for each part family member that can be determined by using historical
data, specification, or prior experience on the similar parts [1]. Wheeler [5] introduced Z chart (also
called Zed bar chart) for subgroup data when n > 1. The mean values for Z chart is then calculated as:
𝑥̄ 𝑖𝑗 −𝑇𝑗
𝑍𝑖𝑗 = 
̂ x̄ 𝑗
(3)

where 
̂x̄ 𝑗 is an estimate of the population standard deviation of the subgroup mean as:
𝑅𝑗

̂ x̄ 𝑗 = 𝑑2 √ 𝑛
(4)
and the ranges, Rij for the W chart are transformed as follows:
𝑅𝑖𝑗
𝑊𝑖𝑗 = 
̂ x𝑗
(5)

Z chart will have a central line at zero with the upper and lower control limits set at +3 and -3,
respectively. W chart set its central line at d2 as R= 𝑅; upper and lower control limits at d2 + 3d3 and
d2 - 3d3 respectively, where d3 is a statistic constant for subgroup size (n).
The advantages of Z chart and W chart, are that the measured data from multiple product families
can be plotted on a single chart and relatively low average run length (ARL) required in comparison to
x̄ chart and R chart [7]. The disadvantages of Z chart and W chart, are that the sigma is often unknown,
a separate estimate is usually obtained for each product and this method requires sufficient data to
negate the fact that the true parameter values used in the calculation of the control limit estimation [8].
These charts are inefficiently adopted in some T&M processes due to the measured data are affected
by measurement error.

2.2. Guard band approaches

The most immediate approach to minimize the measurement errors is by utilizing guard band to
reduce the width of acceptance limit. Various economic aspects of guard band have been proposed by
several researchers (e.g., [9]) to ensure the acceptable risk decision in the product conformity. Industry
practice allows the installation of guard band, e.g., through Guide to the Expression of Uncertainty in

2
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

Measurement (GUM), published by ISO in 1993 to reduce the width of acceptance limit [10], as an
indirect way to compensate the measurement errors. Incorporation of stringent control as in Z chart
and W chart into guard band could potentially aid to screen for assignable causes and to detect early
quality deterioration in the process.

3. Modified SPC model

3.1. Overview
The model standardizes the observed value in relation to the measurement target (T) and measurement
uncertainty (𝑈). Measurement target is referred as an expected value for a predetermined measurement
characteristic. The target value is calculated based on repeated measurement with series observations
for each test station separately and to be revised when the measurement characteristic changes.
Measurement uncertainty refers to the dispersion of the data that could reasonably be attributed to the
measurement result [11]. Pythagoras's theorem is used to estimate the process standard deviation with
population standard deviation and measurement uncertainty as inputs. S-factor (Sf) is introduced as
separate approaches to the control limits to improve the sensitivity in detecting small shifts. Eight tests
are proposed to interpret the charts using Nelson’s rules.

3.2. Evaluation of measurement uncertainty

Measurement uncertainty will be evaluated using GUM [10], in four steps as follows:

3.2.1. Step 1: Identifying all sources of uncertainty. The first step is to identify the output result, Y,
from N input quantities through a function relation f as in equation (6):
Y = f (X1, X2, ..., XN) (6)
where Xi is an input quantity that can significantly affect the measurement result. An estimate of the
output Y, denoted by y, using input estimates x1, x2, ..., xN, as shown in equation (7):
y = f (x1, x2, ..., xN) (7)
Determining the uncertainty of y requires the uncertainties of all the input estimates xi referred as
standard uncertainties 𝑢(𝑥𝑖 ).

3.2.2. Step 2: Evaluating the standard uncertainty. The value for 𝑢(𝑥𝑖 ) , could be obtained from
either Type A or Type B evaluation. Type A evaluation is characterized by a statistically estimated
sample standard deviation si, and the associated number of degrees of freedom vi. The sample standard
deviation of the mean is computed as:
𝑆𝑖
𝑢(𝑥𝑖 ) = (8)
√𝑟
where r was the number of the independent repeated observations. The r should be large enough to
ensure the estimated value is reliable that the probability distribution often is assumed to be normal
[12].
Type B evaluates a component of measurement uncertainty that has been excluded in Type A
evaluation. In the evaluation, 𝑢(𝑥𝑖 ) refers to an assumed probability distribution based on previous
measurement data, input of experienced personnel, manufacturer’s specification, data provided in
calibration report [13]. Amongst various distributions exist, (e.g., rectangular, triangular, U-shaped,
normal.), rectangular distribution is the most common in the analysis of T&M. The accompanying
standard measurement uncertainty is:
𝑎
𝑢(𝑥𝑖 ) = (9)
√3
where a is the semi-range (or half-width) constant set between the upper and lower limits.

3
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

3.2.3. Step 3: Computing the Combined Standard Uncertainty. Combined standard uncertainty,
denoted by 𝑢c (𝑦) is compiled from individual 𝑢(𝑥𝑖 ), using Summation in Quadrature which is often
called the law of propagation of measurement uncertainty, as given below:
1
𝑢c (𝑦) = [∑𝑛1[𝑐𝑖 𝑢(𝑥𝑖 )]2 ]2 (10)
where ci is the sensitivity coefficient associated with xi. Eq. 10 is valid only if the input quantities Xi
are independent or uncorrected. In most cases, input quantity Xi is uncorrelated, and sensitivity
coefficient can be assumed to be 1 [14].

3.2.4. Step 4: Computing the Expanded Uncertainty. Originated from GUM, expanded uncertainty, U
is calculated by factoring in a coverage factor, k into combined standard uncertainty 𝑢c (𝑦):
𝑈 = 𝑘𝑢c (𝑦) (11)
where k is chosen from the effective degrees of freedom vi of all the uncertainty sources, considering a
specified coverage probability, calculated through Student’s t-distribution. Most commonly, k = 2 is
chosen to give a level of confidence of approximately 95% [15]. It is recommended that the expanded
uncertainty (𝑈) be rounded normally to two significant figures [12].
The doubt of the measurement can be quantified when the true value is within the margin with U
and the confidence level [15]. All measurement uncertainty components regardless of classification
are modeled by probability distributions quantified by variances or standard deviations [12].
Therefore, the process standard deviation contributed by measurement error can be estimated using
equation (12), which is termed standard deviation of the measurement uncertainty and denoted by  ̂u .
𝑈

̂u = 𝑘
(12)

3.3. Standardized measurement target and measurement uncertainty

The measurement uncertainty is dependent on the accuracy of the equipment used in the individual
test station. An equation is formed by extending the equation (3) with mth test station instead of jth
part family member. 𝑍𝑖𝑚 for modified Z chart takes measurement uncertainty into account, as depicted
in equation (13):
√ 𝑛 (𝑥̄ 𝑖𝑚 −𝑇𝑚 )
𝑍𝑖𝑚 = 
̂ p𝑚
(13)
where Tm is an expected value of the measurement target and  ̂p𝑚 is an estimate of the process
standard deviation for each test station. Range for modified W chart are extended from equation (5) as
follows:
𝑅𝑖𝑚
𝑊𝑖𝑚 = 
̂ p𝑚
(14)
The ̂p𝑚 provides the tolerance for the control limits. The model iteratively expands the tolerance
limit from the lowest to highest value of 
̂𝑝𝑚 . For the lowest values, min ̂ p𝑚 = ̂x𝑚 , where 
̂ x𝑚 is
an estimate of the population standard deviation that can be calculated by:
𝑅𝑚

̂ x𝑚 = 𝑑2
(15)

where 𝑅𝑚 is the average range for the mth test station. For the highest values, 
̂ u from equation (12) is
taken into account. Pythagoras's theorem is used to estimate the maximum value of the  ̂ p𝑚 , as given
below:
max  ̂ x𝑚 2 + 
̂ p 𝑚 = √ ̂ u𝑚 2 (16)
The central line and control limits of modified charts are fixed with constant values as applied in
the Z chart and W chart. The charts are interpreted using Nelson’s rules as defined in table 1. Rules 1,

4
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

2, 5, and 6 are to be applied to upper and lower halves of the chart separately. Rules 3, 4, 7, and 8 are
to be applied to whole chart [16]. With this, all rules apply to x̄ chart and Z chart. However, only rule
1, 2, 5 and 6 can be applied to dispersion chart (R chart and W chart) without modification when the
subgroup size is 5 or more which will lead to the symmetric control limits [17].

Table 1. Nelson Rules

Rule Description
Rule 1 One point is more than 3 standard deviations () from the mean.
Rule 2 Nine points in a row are on the same side of the mean.
Rule 3 Six points in a row are continually increasing (or decreasing).
Rule 4 Fourteen points in a row alternate in direction, increasing then decreasing.
Rule 5 Two out of three points in a row are more than 2 from the mean in the same direction.
Rule 6 Four out of five points in a row are more than 1 from the mean in the same direction.
Rule 7 Fifteen points in a row are all within 1 of the mean on either side of the mean.
Eight points in a row exist with none within 1 of the mean and the points are in both
Rule 8
directions from the mean.

3.4. Determining ̂𝑝𝑚 through S-factor

S-factor is an approach to estimate 
̂p𝑚 is through the expanded uncertainty (𝑈𝑚 ) and S-factor (𝑆f𝑚 ),
with the formula set as below:
𝑈𝑚

̂ p𝑚 = 𝑆f𝑚
(17)
S-factor is then delineated as the ratio between the highest value of 
̂p𝑚 and the lowest value of 
̂ p𝑚 ,
is given by:
max 
̂p𝑚
𝑆f𝑚 = min 
̂ p𝑚
(18)

and the 
̂p𝑚 is set equal to min 
̂p𝑚 when the estimated value of the 
̂p𝑚 is equal or less than 
̂ x𝑚 ,
where:
𝑈𝑚
min 𝑆f𝑚 = 
̂ x𝑚
(19)
As shown in figure 1, S-factor is useful to rescales the process standard deviation, and it is
dependent on the ratio between the measurement uncertainty and the population standard deviation,
which is termed as U/kx. For example, the  ̂p𝑚 estimated by S-factor is equal to max  ̂ p𝑚 when
U/kx ratio is equal to 1, and it is equal to min 
̂ p𝑚 when U/kx ratio is equal or less than 0.577.
Finally, the 𝑍𝑖𝑚 values for Z chart are given by:
√𝑛 𝑆f𝑚 (x𝑖𝑚 −𝑇𝑚 )
𝑍𝑖𝑚 = 𝑈𝑚
(20)
and range values for W chart are transformed using the following formula:
𝑆f𝑚 𝑅𝑖𝑚
W𝑖𝑚 = 𝑈𝑚
(21)

3.5. Model implementation

In implementation, the model was embedded in an automated SPC system. The SPC application is
developed using Visual Studio to gather data from test record database and to perform data analysis
using analysis tool developed by Scrucca [18]. The tool uses R package (open-source programming

5
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

language) for quality control charting and violating rules checking. Automated SPC includes four
steps: extracting data, creating analysis files, configuring setting and executing the SPC application.
Variables used in the model (e.g., Tm, 𝑈𝑚 , 𝑆f𝑚 ) are defined and to be configured at setting stage. Two
set of charts are plots, as below：
1. Classical Shewhart control charts (x̄ chart and R chart)
2. Z chart and W chart with  ̂p𝑚 determined through S-factor
Eight tests based on Nelson’s rules interpret the charts. False alarm rate is used to compare the
performance of the proposed models with classical Shewhart control charts. A false alarm will be
counted if the rule test detected when the process behaves as per this normal or in-control behavior.

Figure 1. Estimated process standard deviation with S-factor

4. Verification through case study in real industry

A complete year’s data samples were collected from a product tested at multi-stations in a T&M
manufacturing facility at Bayan Lepas, Penang. The DC vertical gain accuracy (VGA) for oscilloscope
test is selected where it consists of 700 measurement results tested at three stations (WH05, WH06,
and WH07). Two sets of charts are constructed with usually recommended subgroup size of 5 [19].
First six months’ data are used in phase I to construct trail control limits. To produce good results in
practice, three sigma control limits were used so that the probability of type I error is kept at 0.0027
[1]. These control limits are used in Phase II to monitor process outcomes with second half six
months’ data. The implementation and analysis for all test stations are performed using same
procedure. In this section, station WH05 is used to demonstrate the implementation with the proposed
model.

4.1. Classical Shewhart control charts (x̄ chart and R chart)

In figure 2, 5 points and 1 point violate the rules in x̄ chart and R chart respectively. However, in
actuality, the process is said in compliance to industry standard as all reading and its tolerance
coincidentally falls within the acceptance limits with measurement uncertainty guard band as shown in
figure 3. With this, the violated points are considered false alarm, except for the points circled in red
are considered true signal where there is large shift (rule 1) and small sustained shift (rule 6) on July
2015. To solve the false alarm problem, operator may compensate the error through process
adjustment or the control engineer will perform the retrospective analysis to construct a new control
limits.

6
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

Figure 2. Shewhart x̄ chart and R chart for DC VGA test in station WH05

Figure 3. Scatter Plot for DC VGA test in station WH05

4.2. 𝑍 chart and W chart with  ̂𝑝𝑚 determined through through S-factor
First, measurement target (T) needs to be determined. DC VGA test is measured by comparing the
voltage reading between oscilloscope (Scope) and digital multimeter (DMM). Tm value is calculated
from repeated measurement for each station and recompiled in the event of DMM readjustment during
the calibration or DMM replacement with another unit. Five input quantities (Xi) affecting the
measurement result (Y) are identified, as shown in table 2. They undergo either Type A or Type B
evaluation to obtain the standard uncertainties 𝑢(𝑥𝑖 ). X1 is obtained from 15 repeated observations, the
input estimates x1 is evaluated using Type A. Other input quantities (X2 to X5) are the equipment
specification and the resolution of the reported measured value, the input estimates x2 to x5 are
evaluated using Type B with rectangular distribution.
The combined standard uncertainty 𝑢c (𝑦) is computed from individual 𝑢(𝑥𝑖 ) using equation (10):
1
𝑢c (𝑦) = [[𝑐1 𝑢(𝑥1 )]2 + [𝑐2 𝑢(𝑥2 )]2 + [𝑐3 𝑢(𝑥3 )]2 + [𝑐4 𝑢(𝑥4 )]2 + [𝑐5 𝑢(𝑥5 )]2 ]2
𝑢c (𝑦) = 0.638 mV

7
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

The expanded uncertainty (𝑈𝑚 ) is calculated with the coverage factor km = 2 and rounded to two
significant figures.
𝑈𝑚 = 2𝑢c (𝑦) = 1.3 mV

Table 2. Evaluation of standard uncertainties

Xi Contributors Type vi ± Limits Units Distribution Ci 𝒖(𝒙𝒊 )
X1 Repeatability tests A 14 3.21E-04 mV Sigma 1 3.21E-04
X2 Voltage accuracy from B  3.15E-05 mV Rectangular 1 1.82E-05
DMM
X3 Resolution error of DMM B  5.00E-03 mV Rectangular 1 2.89E-03
X4 Vertical resolution error of B  7.81E-01 mV Rectangular 1 4.51E-01
Scope with input signal
X5 Vertical resolution error of B  7.81E-01 mV Rectangular 1 4.51E-01
Scope without input signal

With the estimated values of  ̂x𝑚 and ̂u𝑚 , the max ̂ p𝑚 is computed using Pythagoras's theorem
in equation (16). S-factor (𝑆f𝑚 ) is calculated with a ratio between max ̂p𝑚 and min 
̂p𝑚 , as shown in
table 3.
Table 3. Estimation of  ̂p𝑚 with S–factor
Station 𝑼𝒎 
̂ 𝐱𝒎 
̂ 𝐱𝒖 Min  ̂𝐩𝒎 Max  ̂𝐩𝒎 𝑺𝐟𝒎
WH05 1.30E-03 5.63E-04 6.50E-04 5.63E-04 8.60E-04 1.527
WH06 1.30E-03 5.61E-04 6.50E-04 5.61E-04 8.58E-04 1.531
WH07 1.30E-03 6.06E-04 6.50E-04 6.06E-04 8.89E-04 1.466

The transformation expression in equation (20) and equation (21) are applied to plot Z chart and W
chart respectively as shown in figure 4. The results showed that there were no false alarm points found
and capable to detect the large shift from mean. As of now, the implementation is completed for
station WH05; similar process will be repeated for other stations WH06 and WH07.

Figure 4. Modified Z chart and W chart for DC VGA test in station WH05 with S-factor

5. Result and Discussion

The summarized results of the case study are recorded in table 4. At this case, the U/kx ratio is at 1.1,
the measurement error marginally affects the process standard deviation. From the analysis results
from Shewhart x̄ chart, all test stations had many points exceeding the control limits and some
abnormal patterns are falsely detected. These false alarms caused many unnecessary process

8
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

adjustment and loss of confidence in SPC. With the S-factor approach, the analysis results showed that
the model performed very well in both Z chart and W chart with zero false alarms in all stations. The
model reduced false alarms rate by up to 13% in comparison to the classical Shewhart x̄ chart. Based
on the results of observations, all test stations are in normal and controlled behavior, except for station
WH05 where the shift occurs in the process. In this case study, all methods were able to detect these
assignable causes. Results of the case study proved that the model with S-factor is efficient and
effective for T&M process.

Table 4. Summary result of the case study

6. Conclusions
This paper proposed a modified SPC model combining modified guard band and control charts (Z
chart and W chart) to address issues caused by short production runs and measurement errors. The
model utilized the measurement target and measurement uncertainty to standardize the observed value.
S-factor determines the estimation of the process standard deviation for control limits tolerance. The
effectiveness of proposed model was demonstrated by a case study in real industry and it shows that
the measurement error marginally affected the process standard deviation. The results revealed that the
model significantly reduced false alarm rate compared to the one using classical Shewhart control
charts. Moreover, the results proved that the model using the S-factor can rescale the process standard
deviation in an efficient and effective manner. The proposed model using the modified control charts
is practical for T&M manufacturing to eliminate false alarm, without sacrificing sensitivity to level
shifts.

References
[1] Montgomery D C 2013 Statistical quality control: A modern introduction 7th ed. (Singapore:
John Wiley & Sons)
[2] Bothe D R 1989 A powerful new control chart for job shops 43rd Annual Quality Congress
Transactions May 1989 (Toronto) (ASQC) pp 265-270
[3] Costa A F B and Castagliola P 2011 Effect of measurement error and autocorrelation on the X̄
chart Journal of Applied Statistics 38(4) 611-673
[4] Barrera J 2012 Analysis of some statistics used in short runs charts Proc. of 15th International
Conference on Experimental Mechanics 22-27 July 2012 (Porto/Portugal) (ICEM15) pp 1-2

9
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012009 doi:10.1088/1757-899X/328/1/012009
1234567890‘’“”

[5] Wheeler D 1991 Short Run SPC SPC press

[6] Farnum N 1992 Control charts for short runs: nonconstant process and measurement error
Journal of Quality Technology 24(3) 138-144
[7] Moore S S and Murphy E 2013 Process visualization in medical device manufacture: an
adaptation of short run SPC techniques Quality Engineering 25(3) 247-265
[8] Del Castillo E, Grayson J M, Montgomery D C and Runger G C 1996 A review of statistical
process control for short run manufacturing systems Communications in Statistics - Theory
and Methods 25(11) 2723–2737
[9] Pou J M and Leblond L 2015 Control of customer and supplier risks by the guardband method
International Journal of Metrology and Quality Engineering 6 205
[10] JCGM 100: 2008 Evaluation of measurement data– Guide to the expression of uncertainty in
measurement (GUM 1995 with minor corrections) (Joint Committee for Guides in
Metrology)
[11] JCGM 200: 2012 International Vocabulary of Metrology – Basic and General Concepts and
Associated Terms (VIM 3rd ed. ) (Joint Committee for Guides in Metrology)
[12] UKAS 2007 The Expression of Uncertainty and Confidence in Measurement M3003 2nd ed.
(UK: United Kingdom Accreditation Service)
[13] Taylor B N and Kuyatt C E 1994 Guidelines for Evaluating and Expressing the Uncertainty of
NIST Measurement Results (MD: National Institute of Standards and Technology, USA)
[14] Instone I 1996 Simplified method for assessing uncertainties in a commercial production
environment Teste and measurement Metrology Forum (London) pp 1–7
[15] Bell S 2001 Measurement Good Practice Guide No 11 (Issue 2): A Beginner's Guide to
Uncertainty of Measurement (UK: National Physical Laboratory)
[16] Nelson L S 1984 The Shewhart control chart-tests for special causes Journal of Quality
Technology 16(4) 237–239
[17] Noskievicova D 2013 Complex control chart interpretation International Journal of
Engineering Business Management 5 (13:2013) 1-7
[18] Scrucca L 2004 qcc: An R package for quality control charting and statistical process control R
News 4(1) 11-17
[19] Reynolds M R and Stoumbos Z G 2004 Should observations be grouped for effective process
monitoring Journal of Quality Technology 36 343–366

10