Computer Modeling in

Tech Science Press

Engineering & Sciences

DOI: 10.32604/cmes.2022.019199 Published Online: 18 January 2022

ARTICLE

New Hybrid EWMA Charts for Efficient Process Dispersion

Monitoring with Application in Automobile Industry

Xuechen Liu1 , Majid Khan2 , Zahid Rasheed3 , Syed Masroor Anwar4,* and Muhammad Arslan5
1
Faculty of Economics, Taiyuan Normal University, Taiyuan, 030619, China
2
Department of Mathematics and Statistics, Riphah International University, Islamabad, 44000, Pakistan
3
Department of Mathematics, Women University of Azad Jammu and Kashmir, Bagh, 12500, Pakistan
4
Department of Statistics, University of Azad Jammu and Kashmir, Muzaffarabad, 13100, Pakistan
5
School of Statistics, Shanxi University of Finance and Economics, Taiyuan, 030619, China
*
Corresponding Author: Syed Masroor Anwar. Email: [email protected]
Received: 08 September 2021 Accepted: 09 November 2021

ABSTRACT
The EWMA charts are the well-known memory-type charts used for monitoring the small-to-intermediate shifts in
the process parameters (location and/or dispersion). The hybrid EWMA (HEWMA) charts are enhanced version
of the EWMA charts, which effectively monitor the process parameters. This paper aims to develop two new upper-
sided HEWMA charts for monitoring shifts in process variance, i.e., HEWMA1 and HEWMA2 charts. The design
structures of the proposed HEWMA1 and HEWMA2 charts are based on the concept of integrating the features of
two EWMA charts. The HEWMA1 and HEWMA2 charts plotting statistics are developed using one EWMA statis-
tic as input for the other EWMA statistic. A Monte Carlo simulations method is used as a computational technique
to determine the numerical results for the performance characteristics, such as average run length (ARL), median
run length, and standard deviation run length (SDRL) for assessing the performance of the proposed HEWMA1
and HEWMA2 charts. In addition, to evaluate the overall performance of the proposed HEWMA1 and HEWMA2
charts, other numerical measures consisting of the extra quadratic loss (EQL), relative average run length (RARL),
and performance comparison index (PCI) are also computed. The proposed HEWMA1 and HEWMA2 charts are
compared to some existing charts, such as CH, CEWMA, HEWMA, AEWMA HHW1, HHW2, AIB-EWMA-I, and
AIB-EWMA-II charts, on the basis aforementioned numerical measures. The comparison reveals that the proposed
HEWMA1 and HEWMA2 charts achieve better detection ability against the existing charts. In the end, a real-life
data application is also provided to enhance the implementation of the proposed HEWMA1 and HEWMA2 charts
practically.

KEYWORDS
Average run length; extra quadratic loss; memory-type charts; Monte Carlo simulations; smoothing parameter

This work is licensed under a Creative Commons Attribution 4.0 International License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
2 CMES, 2022

1 Introduction
Control charts are the essential tools of the statistical process monitoring (SPM) toolkit, used
to detect the shifts in manufacturing and production processes parameter(s). The control charts are
generally classified into memory-type and memoryless-type charts [1]. The memoryless-type charts
are used only for current information of the process, while the memory-type charts are based on
both current and previous information of the process. The basic memoryless-type charts are the
Shewhart charts, like the Shewhart X , R, and S2 charts, etc. [2]. The Shewhart charts are simple
and easy to apply; however, they are only efficient for the cases where large shifts occur in the
process parameter(s). On the contrary, the memory-type charts, such as the exponentially weighted
moving average (EWMA) and cumulative sum (CUSUM) charts are sensitive in monitoring small-
to-intermediate shifts in the process parameter(s).
Roberts [3] was the first to introduce the classical EWMA chart for monitoring the mean
level of the process. Later, Hunter et al. [4–7] further investigated the various EWMA-type charts
in order to facilitate the application of the classical EWMA chart in mean process monitoring.
Recent studies have shown that EWMA-type charts are the most useful tools for researchers. For
example, Haq [8], Abbas et al. [9], Ali et al. [10], Tang et al. [11], Haq [12], Rasheed et al. [13],
Rasheed et al. [14], etc., are the few recent references in this regard.
In general, most manufacturing and production processes have a shift in the mean level;
however, the process variance (or standard deviation) may be shifted from the target in many
practical situations. Domangue et al. [15] suggested that monitoring an increase in process variance
is more important for the processes. Although the EWMA-types charts are primarily used in
mean process monitoring; however, few works address the variance monitoring via these charts.
For example, Crowder et al. [16] used the logarithmic transformation to the sample variance
S2 in order to develop the EWMA chart (also known as the CH chart) for monitoring the
process standard deviation. Similarly, Shu et al. [17] suggested the EWMA chart, denoted as the
NEWMA chart, efficiently monitors process variance compared to the CH chart. Correspondingly,
Huwang et al. [18] developed the EWMA-type charts for detecting shifts in the process variance
and demonstrated that their control charts outperform the CH and NEWMA charts. Equally,
Castagliola [19] used the three parameters logarithmic transformation of S2 and proposed the
bilateral EWMA chart to monitor the process variance shifts. Besides, Chang et al. [20] designed
the optimal EWMA chart in order to monitor the process variance shifts. In addition, Razmy [21]
offered the EWMA chart the monitors the standardized process variance. Furthermore, Haq [22]
proposed two auxiliary information-based charts, symbolized by AIBEWMA1 and AIBEWMA2
charts, that monitor the process variance efficiently. Also, Ali et al. [23] suggested the generally
weighted moving average (GWMA) and hybrid EWMA (HEWMA) chart to monitor process
variance changes. Both GWMA and HEWMA perform better than classical memory charts. Other
studies based on the EWMA-type charts for tracking the process variance are provided by Saghir
et al. [24], Zaman et al. [25], Riaz et al. [26] and Chatterjee et al. [27], etc.
The use of hybrid charts enhances the efficiency of traditional charts. For example, Haq [8]
and Haq [28] proposed the HEWMA chart to monitor the mean level of the process. Later
on, numerous authors used the HEWMA charts in different process monitoring schemes. The
HEWMA charts are more efficient than the classical EWMA and CUSUM charts in terms of
small to moderate shifts monitoring. For example, Aslam et al. [29] introduced the HEWMA
chart to monitor the mean level of the process under repetitive sampling. Similarly, Aslam
et al. [30] monitored the COM-Poisson process by designing the HEWMA chart. Equally, Aslam
et al. [31] developed the mixed chart, named the HEWMA-CUSUM chart, for the Weibull process
CMES, 2022 3

monitoring. Correspondingly, Noor-ul-Amin et al. [32] recommended the HEWMA chart for
Phase-II mean monitoring, based on the auxiliary information. Besides, Aslam et al. [33] suggested
the HEWMA-p chart to monitor the variance of the non-normal process. Also, Noor et al. [34]
constructed the Bayesian HEWMA chart, using two loss functions, to monitor the mean level of
the normal process. The other studies about the HEWMA chart are offered by Asif et al. [35],
Noor-Ul-Amin et al. [36], etc.
The majority of manufacturing and service processes are affected by the gradual increase
in process variance. The increase in the process variance indicates a deterioration in the pro-
cess performance. This study’s first and most important goal is to propose efficient charts with
effective shifts detection ability in monitoring an increase in the process variances. So, motivated
by Crowder et al. [16], Castagliola [19], and Haq [8], this study proposes two new HEWMA
charts to monitor the increasing shifts in the process variance. The proposed charts are known
as HEWMA1 and HEWMA2 charts. The design structure of the HEWMA1 chart uses the CH
statistic as the input for the HEWMA1 statistic. In the same lines, the CEWMA statistic is
considered an input to the HEWMA2 statistic to formulate the HEWMA2 chart. The Monte
Carlo simulations are employed to compute the numerical results associated with average run
length (ARL), standard deviation run length (SDRL), extra quadratic loss (EQL), relative average
run length (RARL), and performance comparison index (PCI) for the proposed HEWMA1 and
HEWMA2 charts. Based on these measures, the proposed HEWMA1 and HEWMA2 charts are
compared to the existing CH, CEWMA, HEWMA, AEWMA, HHW1, HHW2, AIBEWMA1,
and AIBEWMA2 charts. The comparison shows that the proposed HEWMA1 and HEWMA2
charts have better detection ability to detect the shift in the process variance. Finally, two appli-
cations of the proposed HEWMA1 and HEWMA2 charts are provided, one with simulated data
and the other with real-life data, to aid in the comparison of the proposed charts.
The remainder of the article is set out in the following way: Section 2 presents the exist-
ing methods. Likewise, Section 3 lays out the methodologies and formulation of the proposed
HEWMA1 and HEWMA2 charts. In addition, the performance evaluation measures and simu-
lation study are included in Section 4. Furthermore, Section 5 consists of the comparison and
performance analysis of the proposed HEWMA1 and HEWMA2 charts against some existing
charts. Section 6 offers a real-life data application to enhance the performance of the proposed
HEWMA1 and HEWMA2 charts. The last section addresses the concluding remarks.

2 Existing Schemes
This section defines the process variable in Subsection 2.1. Similarly, Subsection 2.2 explains
the details about the transformations to the sample variance. In addition, Subsections 2.3 to 2.6
provide the design and formulation of the CH, CEWMA, HHW2, and HEWMA charts for
monitoring process variance, respectively.

2.1 Process Variable

Assuming, at the time t = 1, 2, . . ., there are X1t , X2t , . . . , Xit , . . . , Xnt ; i.e., n independent
 iden-
tically normal random variables with mean and variance μt and σt , i.e., Xit ∼N μt , σt for2 2

i = 1, 2, . . . , n. As the only concern is to detect the increasing changes in σt2 , so the mean level
of the process is assumed to be IC, i.e., μt = μ. Suppose the underlying process variance remains
IC for a particular time, i.e., σt2 = σ02 for t < t0 , then the process goes in OOC state, i.e., σt2 = σ02
for t ≥ t0 . Let δt denotes the size of shifts in σt2 then it can be defined as a ratio IC process
standard deviation to OOC process standard deviation, i.e., δt = σσt . So, δt = 1 whenever the
0
4 CMES, 2022


underlying process is IC, and δt = 1 in the case of the OOC process. If X t = 1n ni=1 Xit be the
1 n
 2
sample mean and St2 = n−1 i=1 Xit − X t denotes the sample variance of the tth subgroup,
respectively, then for IC process St2 follows a chi-square distribution with n −1 degrees of freedom,
σ2
i.e., St2 ∼ n−1
0
χ(n−1)
2 .

2.2 Transformation
In order to implement the EWMA-type charts, the assumption of normality is required for
the plotting statistic of the charts. However, because the sample variance St2 has a chi-square
distribution, so it is not an acceptable statistic for the design structure of the EWMA-type charts.
In order to cope with this issue, a few transformations are available in the literature, given as
follows.

2.2.1 Transformation-I
Let Wt denote a log transformation of St2 , defined as
 
Wt = ln St2 /σ02 , (1)

where the ratio St2 /σ02 is follows the gamma distribution with parameter (n − 1)/2 and 2δt2 /(n − 1).
According to Lawless [37], Wt is a log-gamma random   variable and it has a approximate normal
distribution, i.e., Wt  N(μW , σW 2 ), where μ = ln δ 2 − 1 −
W t n−1
1
2 +
2
4 and σW = n−1 +
2 2
3(n−1) 15(n−1)
2
+ 4
+ 16
.
(n−1)2 3(n−1)3 15(n−1)5

2.2.2 Transformation-II
Castagliola [19] recommended the three-parameter logarithmic transformation of St2 into a
new variable, given as
 
Tt = aT + bT ln St2 + cT , (2)

where aT , bT and cT are the three constants of the transformation, defined as; aT = AT −
2BT ln(σ ), bT = BT , cT = CT σ , respectively. The values of AT , BT , CT , μT and σT2 based on
the sample size n (n = 3, 4, 5, 10, 15) are given by Castagliola [19]. The statistic
 Tt , in this case, is
approximately normally distributed with μT and variance σT2 , i.e., Tt ∼N μT , σT2 .

2.2.3 Transformation-III
In order to detect the shifts in the process variance, Quesenberry [38] suggested another
transformation defined as
  
Mt = Φ−1 F χt2 ; ν , (3)

(n−1)St2
where χt2 = , is chi-square variable with ν = n − 1 degrees of freedom, i.e., χt2 ∼χ(ν)
2 . The
σ02
F (·; ν) is the distribution function (DF) of the chi-square variable, while Φ−1 (·) denotes the
inverse DF of the standard normal variable. In this case, the statistic Mt follows a standard
normal distribution, i.e., Mt ∼N (0, 1).
CMES, 2022 5

2.3 CH Chart
Crowder et al. [16] introduced the CH chart, which monitored the process variance shifts. Let
{Qt } for t ≥ 1 be the sequence of IID random variable, defined on the sequence {Wt }, where Wt
is defined by Eq. (1), then using the recurrence relationship, the CH plotting statistic Qt can be
given as
Qt = (1 − λ1 ) Qt−1 + λ1 Wt , (4)
where λ1 is known as the smoothing parameter. The CH statistic mean and variance are,
respectively, given as
λ1
E (Qt ) = μW and Var (Qt ) = [1 − (1 − λ1 )2t ]σW
2
.
2 − λ1
λ1
In the case of large t, the variance of Qt is reduced to; Var (Qt ) = 1−λ1 σW .
2 In order to detect
the gradual rise in the process variance, let Q+
t be the charting statistic for the CH chart is given
by
Q+
t = max(Qt , 0). (5)

The initial value of and Q+ +

t is set on 0, i.e., Q0 = 0. The upper control limit for the CH chart
is denoted by UCL+ (CH) and can be defined by

λ1
UCL+ +
(CH) = LCH 2 − λ σW . (6)
1

where L+
CH is the CH chart width coefficients and can be computed so that IC ARL (ARL0 ) is
approximately equal to the desired value. The CH chart detects the upward shifts in the process
whenever Q+ +
t ≥ UCL(CH) .

2.4 CEWMA Chart

Castagliola [19] proposed the EWMA chart (also known as S2 -EWMA chart) to monitor the
shifts in the process variance. Hereafter the S2 -EWMA chart is labeled as the CEWMA chart.
The CEWMA chart used the three parameters logarithmic transformation to S2 to obtain the
approximate normality for the plotting statistic. Let {Zt } be the CEWMA sequence, based on
another sequence {Tt } for t ≥ 1, where Tt is defined by Eq. (2), then the charting statistic of the
CEWMA chart is given by
Zt = (1 − λ1 ) Zt−1 + λ1 Tt , (7)

The initial value of Zt is denoted by Z0 and can be is defined as

Z0 = AT + BT ln (1 + CT ) . (8)

The Z0 values for various n can be taken from [19]. The Z0 values are close to 0, so one
can replace them with 0. Here, the plotting statistic Zt is an approximate normal distributed vari-
 
λ1 λ1
able having mean μT and variance 2−λ 1
[1 − (1 − λ1 )2t 2
]σ T , i.e., Z t ∼N μ ,
T 2−λ1 [1 − (1 − λ1 ) 2t 2
]σ T .

However, for a large value of t, the factor [1 − (1 − λ1 )2t ] tends to 1, and in this case
6 CMES, 2022

 
λ1
Zt ∼N μT , 2−λ σT2 . If UCL+
(CEWMA) denotes the upper control limit for the CEWMA chart, then
1
it can be defined as

λ1
UCL+ +
(CEWMA) = μT + LCEWMA σT , (9)
2 − λ1

where L+
CEWMA is called the width coefficient of the CEWMA chart. The CEWMA chart detects
OOC signals with increasing shift whenever Zt ≥ UCL+
(CEWMA) .

2.5 HHW2 Chart

Huwang et al. [18] proposed the EWMA chart to monitor the process variance, denoted as
the HHW2 chart. Let {Rt }, for t ≥ 1 be the HHW2 sequence that based on the IID sequence
of random variable {Mt }, where Mt is defined in Eq. (3), then the HHW2 statistic based on the
sequence {Rt } is defined as
Rt = (1 − λ1 ) Rt−1 + λ1 Mt . (10)

The initial value of and Rt is set on 0, i.e., R0 = 0. The plotting statistic Rt has a normal
  
λ1 λ1
distribution with mean zero, and variance 2−λ 1
1 − (1 − λ 1 )2t
, i.e., R t ∼N 0, 2−λ1 [1 − (1 − λ1 )2t
] .
The time-dependent control limits, UCLt(HHW 2) for the HHW2 chart can be defined as

+ + λ1 
UCLt(HHW 2) = LHHW 2 1 − (1 − λ1 )2t . (11)
2 − λ1

In case of large t values, the upper control limit for of the HHW2 chart is denoted as
UCL+
(HHW 2) and can be given as

λ1
UCL+ +
(HHW 2) = LHHW 2 2 − λ . (12)
1

The L+HHW 2 is the HHW2 chart width coefficients. The HHW2 chart detects OOC signals
whenever Rt fall above the control limits specified in Eq. (12).

2.6 HEWMA Chart

Ali et al. [23] followed the idea of [8] and designed the HEWMA chart for process variance.
Let the IID {Ht } for t ≥ 1 is known as the HEWMA sequence, which is based on the {Rt }, then
the statistic Ht for the HEWMA chart is given as
Ht = (1 − λ2 ) Ht−1 + λ2 Rt , (13)
where Rt is defined by Eq. (10) and λ2 is also a smoothing constant, such that λ2 = λ1 . The initial
values of Ht are set to 0, i.e., H0 = 0. The mean E (Ht ) = 0, and the variance of Ht is given as
CMES, 2022 7

⎡ 

  2 (1 − λ )2 1 − (1 − λ )2t  ⎤
λ1 λ 2 k k 2 (1 − λ1 ) (1 − λ2 ) 1 − (1 − λ1 ) (1 − λ2 )
t t
Var(Ht ) = ⎣ − ⎦.
λ 1 − λ2 1 − (1 − λ k )2 1 − (1 − λ 1 ) (1 − λ 2 )
k=1

(14)

The control limit UCLt(HEWMA) for the HEWMA chart, based on Eq. (14) is defined by
UCL+
t(HEWMA)

= L+
HEWMA
 ⎡ 


 2 (1 − λ )2 1 − (1 − λ )2t  ⎤
 λ1 λ2 k k 2 (1 − λ1 ) (1 − λ2 ) 1 − (1 − λ1 )t (1 − λ2 )t
× ⎣ − ⎦.
λ 1 − λ2 1 − (1 − λk )2 1 − (1 − λ1 ) (1 − λ2 )
k=1
(15)

When t gets larger, the control limits defined by Eq. (15) is reduced to UCL+ (HEWMA) and
given as
  2 


 λ1 λ2   (1 − λ )2
2 (1 − λ ) (1 − λ )
UCL(HEWMA) = LHEWMA 
+ + k 1 2
− . (16)
λ 1 − λ2 1 − (1 − λk )2 1 − (1 − λ1 ) (1 − λ2 )
k=1

where L+HEWMA is the width coefficient for the HEWMA chart. The HEWMA chart triggers OOC
signals whenever Ht ≥ UCL+(HEWMA) .

3 Proposed Methods
Haq [8] suggested the HEWMA chart to monitor the process mean. Similarly, Ali et al. [23]
presented the HEWMA chart’s design structure for tracking the process variance. Following
Haq [8], the HEWMA1 and HEWMA2 charts can be developed using Transformations I and II,
respectively. These charts detect increasing shifts in process variance. The methodologies and con-
struction of the HEWMA1 and HEWMA2 charts are, respectively, presented in Subsections 3.1
and 3.2.

3.1 HEWMA1 Chart

The design structure for the HEWMA1 chart can be constructed using the CH statistic as
input for the HEWMA1 statistic. Let defined the sequence of IID random variable, say {Ut } for
t ≥ 1, based on the CH sequence {Qt } then the HEWMA1 statistic Ut can be defined by the
relation given as
Ut = (1 − λ2 ) Ut−1 + λ2 Qt , (17)
where Qt is the CH statistic defined by Eq. (4). The starting value of Ut is equal to 0, i.e., U0 = 0.
The mean of Ut is E (Ut ) = 0 and its variance, for the case of very large t, is defined as

 
2

λ1 λ 2 (1 − λk )2 2 (1 − λ1 ) (1 − λ2 )
Var(Ut ) = − σW
2
. (18)
λ 1 − λ2 1 − (1 − λk )2 1 − (1 − λ1 ) (1 − λ2 )
k=1
8 CMES, 2022

In order to monitor the increasing shift in the process, the HEWMA1 statistic is defined as
Ut+ = max(Ut , 0). (19)

The initial value of Ut+ is set on 0, i.e., U0+ = 0. The control limit for upper sided HEWMA1
is given as
  2 


 λ1 λ2   (1 − λ )2
2 (1 − λ ) (1 − λ )
UCL+ +  k 1 2
(HEWMA1) = LHEWMA1 λ 1 − λ2
− σW . (20)
k=1
1 − (1 − λk )2 1 − (1 − λ1 ) (1 − λ2 )

where L+HEWMA1 is the width coefficient for the HEWMA1 chart. The HEWMA1 chart detects
OOC signals whenever Ut+ ≥ UCL+ (HEWMA1) . Similarly, for monitoring the gradual decrease in the
variance, the HEWMA1 statistic is given by
Ut− = min(Ut , 0). (21)

The initial value of Ut− is denoted by U0− set on 0, i.e., U0− = 0. The control limit for lower
sided HEWMA1 is given as
  2 


 λ1 λ 2   (1 − λ )2
2 (1 − λ ) (1 − λ )
LCL− −  k 1 2
(HEWMA1) = LHEWMA1 λ 1 − λ2
− σW . (22)
1 − (1 − λk )2 1 − (1 − λ1 ) (1 − λ2 )
k=1

The lower-sided HEWMA1 chart detects OOC signals whenever Ut− ≤ LCL− (HEWMA1) . The
control limits defined in Eqs. (20) and (22) are called the HEWMA1 upper and lower control
limits, respectively. However, the HEWMA1 two-sided control limits are given as
  2  ⎫


 λ1 λ2   (1 − λk )2
2 (1 − λ1 ) (1 − λ2 )
⎪
⎪
⎪
UCL(HEWMA1) = LHEWMA1  − σ ⎪
⎪
λ 1 − λ2 − (1 − λ ) (1 − λ )
W ⎪
⎬
1 − (1 − λ k )2 1 1 2
 k=1
 2  ⎪. (23)


 λ1 λ 2   (1 − λk )2
2 (1 − λ1 ) (1 − λ2 ) ⎪
⎪
LCL(HEWMA1) = −LHEWMA1  − σW ⎪
⎪
⎪
λ 1 − λ2 1 − (1 − λk ) 2 1 − (1 − λ1 ) (1 − λ2 ) ⎭
k=1

The two-sided HEWMA1 chart detects OOC signals whenever Ut ≥ UCL(HEWMA1) or Ut ≤

LCL(HEWMA1) .

3.2 HEWMA2 Chart

In order to formulate the design of the HEWMA2 chart, the charting statistic of the
CEWMA chart can be used as an input for the HEWMA2 statistic. Let {Vt } for t ≥ 1 be the
HEWMA2 sequence, then the HEWMA2 chart statistic is Vt , and it can be defined by
Vt = (1 − λ2 ) Vt−1 + λ2 Zt , (24)
where Zt is the HHW2 plotting statistic defined by Eq. (7). The starting values of Vt is equal to
Z0 defined by Eq. (8). The expected value for the statistic Vt is E (Vt ) = μT and its variance is
given as
CMES, 2022 9


λ1 λ2
Var (Vt ) =
λ 1 − λ2
⎡ 
 ⎤
2 (1 − λ k )2
1 − (1 − λ k )2t
2 (1 − λ1 ) (1 − λ2 ) 1 − (1 − λ1 )t (1 − λ2 )t
×⎣ − ⎦ σT2 . (25)
1 − (1 − λ k)
2 1 − (1 − λ 1 ) (1 − λ 2 )
k=1

The HEWMA2 control limits UCLt(HEWMA2) and LCLt(HEWMA2) are given as

⎫
UCLt(HEWMA2) ⎪
⎪
⎪
⎪
= μT + L(HEWMA2) ⎪
⎪
 ⎪
⎪
 ⎡  ⎤ ⎪
⎪

 2 (1 − λk ) 1 − (1 − λk )
2 2t   ⎪
⎪
 λ1 λ 2 2 (1 − λ1 ) (1 − λ2 ) 1 − (1 − λ1 ) (1 − λ2 )
t t ⎪
⎪
× ⎣ − ⎦σT ⎪
⎪
⎪
λ 1 − λ2 1 − (1 − λ )2 1 − (1 − λ1 ) (1 − λ2 ) ⎪
⎬
k=1 k
.
LCLt(HEWMA2) ⎪
⎪
⎪
⎪
= μT − L(HEWMA2) ⎪
⎪
 ⎪
⎪
 ⎡  ⎤ ⎪
⎪

    ⎪
⎪
 λ1 λ 2
2 (1 − λ ) 1 − (1 − λ )
2 2t
(1 − λ ) (1 − λ ) − (1 − λ ) t
(1 − λ )t ⎪
⎪
 ⎣
k k 2 1 2 1 1 2
⎦ ⎪
⎪
× − σ T ⎪
⎪
λ 1 − λ2 1 − (1 − λk )2 1 − (1 − λ 1 ) (1 − λ 2 ) ⎭
k=1

(26)

The control limits defined in Eq. (26) are known as the two-sided time-dependent control
limits; however, in the case of large t values, the two-sided fixed HEWMA2 control limits are
defined as


 2  ⎫
 λ1 λ2   (1 − λk ) 2
2 (1 − λ1 ) (1 − λ2 )
⎪
⎪
⎪
UCL(HEWMA2) = μT + L(HEWMA2)  − σ ⎪
⎪
λ 1 − λ2 − (1 − λ ) (1 − λ )
T ⎪
⎬
1 − (1 − λ k )2 1 1 2
 k=1
 2  ⎪.


 λ1 λ 2   (1 − λk )2
2 (1 − λ1 ) (1 − λ2 ) ⎪
⎪
LCL(HEWMA2) = μT − L(HEWMA2)  − σ ⎪
⎪
λ 1 − λ2 1 − (1 − λ1 ) (1 − λ2 )
T ⎪
⎭
1 − (1 − λk )
k=1
2

(27)

In this case, The HEWMA2 chart triggers OOC signals whenever Vt ≥ UCL(HEWMA2) or Vt ≤
LCL(HEWMA2) . The control limits specified in Eq. (27) are the two-sided control limits; however,
the HEWMA2 upper control limit is defined as
  2 


 λ1 λ2   (1 − λ )2
2 (1 − λ ) (1 − λ )
UCL(HEWMA2) = μT + LHEWMA2 
+ + k 1 2
− σT . (28)
λ 1 − λ2 1 − (1 − λk )2 1 − (1 − λ1 ) (1 − λ2 )
k=1

The L+HAEWMA2 is known as the chart constants. The HEWMA2 chart detects OOC signals
with increasing shift whenever Vt ≥ UCL+
(HAEWMA2) .
10 CMES, 2022

4 Performance Analysis and Simulation Study

This section defines the performance evaluation measures, such as average run length in
Subsection 4.1 and overall performance measures in Subsection 4.2. Similarly, the simulation
study for the proposed HEWMA1 and HEWMA2 charts is designed in Subsection 4.3. Like-
wise, Subsection 4.4 provides the choices design parameters for the proposed HEWMA1 and
HEWMA2 charts.

4.1 Average Run Length

The most popular and commonly used performance evaluation measures are the ARL and
SDRL measures. The ARL can be defined as the average number of sample points until a chart
indicates the OOC signal [2]. The ARL is further categorized as IC ARL (ARL0 ), and OOC
ARL (ARL1 ). When a process is working in an IC state, the ARL0 should be as large enough to
prevent the frequent false alarms, while the ARL1 should be smaller so that the shift is detected
quickly [39]. A chart with smaller ARL1 is preferred over the competing charts at a prespecified
ARL0 [40].

4.2 Overall Performance Evaluation Measures

The ARL and SDRL measures evaluate the performance of the charts on a single specified
shift. However, sometimes the researcher may want to investigate the charts performances for the
entire range of shifts, i.e., δmin < δ < δmax . For this purpose, the other performance measures, such
as extra quadratic loss (EQL), relative average run length (RARL), and performance comparison
index (PCI), are used. The details on the EQL, RARL, and PCI measures are provided in the
following subsections.

4.2.1 Extra Quadratic Loss

The EQL can be considered as a weighted average of ARL defined over the range of shifts
δmin to δmax , using δ 2 as a weight. Symbolically, the EQL can be defined as
 δmax
−1
EQL = (δmax − δmin ) δ 2 ARL (δ) dδ.
δmin

where ARL(δ) is the ARL at specific shift δ and δmin and δmax are the minimum and maximum
shift values, respectively. A chart with a low EQL value is considered to have a better overall
detection ability [41].

4.2.2 Relative Average Run Length

Like the EQL measure, the RARL also evaluates the overall performance of the charts. The
RARL mathematically can be defined as
 δmax
−1 ARL(δ)
RARL = (δmax − δmin ) dδ.
δmin ARLbenchmark (δ)

where ARLbenchmark (δ) is the ARL value for benchmark chart at shift δ. A chart with a smaller
ARL at specific δ is known as a benchmark chart. The RARL value for the benchmark chart
is always equal to 1. The benchmark chart is considered superior to the competing chart if
RARL > 1 [42].
CMES, 2022 11

4.2.3 Performance Comparison Index

The PCI also assesses the overall performance of the best chart. Ou et al. [43] defined the
PCI as a ratio of EQL of the best chart to the EQL of the benchmark chart. Mathematically, it
can be given by the expression given as
EQL
PCI = .
EQLbenchmark

The PCI value for the benchmark chart is equal to be 1, and for the rest of the charts,
PCI > 1 [44].

4.3 Monte Carlo Simulations

The random sample of size n, i.e., X1t , X2t , . . . , Xit , . . . , Xnt for t > 1, is generated from a
normal distribution under different parameter settings. Domangue et al. [15] suggested that mon-
itoring a gradual rise (process deterioration) in the process variance is more important; therefore,
an upward shift is considered in the process variance. The shift is reflected in the process standard
deviation, i.e., σ = δσ0 , where δ = 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9 and 2.0. The Monte
Carlo simulation approach is utilized as a computational methodology for the numerical results
by designing an algorithm in the statistical package R. At each shift size δ, the simulations are
performed with 20,000 replicates. The simulation algorithm for the HEWMA1 and HEWMA2
charts include the following steps:
(i) Specify sample size n, smoothing parameters   (λ1 , λ2 ) and parameters of the pro-

cess distribution, i.e., for IC process N μ0 , σ02 and for OOC process N μ0 , (δσ0 )2 ,
where δ = σσ0 .
 
(ii) Generate random observations X1t , X2t , . . . , Xnt for t = 1, 2, . . . from N μ0 , σ02 .
(iii) Compute the statistics, Wt and Tt in Eqs. (1) and (2), respectively.
(iv) Using Wt from Tt , compute the EWMA statistics, Qt and Zt in Eqs. (4) and (7),
respectively.
(v) Using Qt and Zt , compute the HEWMA1 statistic Ut+ and HEWMA2 statistic Vt in
Eqs. (19) and (24), respectively.
(vi) Selected L+ +
HEWMA1 and LHEWMA2 for desired ARL0 and compute UCL(HEWMA1) and
+

UCL+
(HEWMA2) in Eqs. (20) and (28), respectively.
(vii) Plot the statistic Ut+ against the UCL+
(HEWMA1) and the statistic Vt against the
UCL+ + + +
(HEWMA2) . If Ut ≥ UCL(HEWMA1) and Vt ≥ UCL(HEWMA2) then record sequence
order called the run length for the HEWMA1 and HEWMA2 charts, respectively.
(viii) Repeat Steps (ii)–(vii) m = 105 times and record run lengths and hence compute the approx-
m

j=1 (RLj −ARL)
m 2
j=1 RLj
imate ARL by ARL = m and approximate SDRL by SDRL = m−1 .
 ARL1 values
(ix) For  generate random observations Xit , X2t , . . . , Xnt for t = 1, 2, . . . from
N μ0 , (δσ0 )2 and repeat the Steps (iii)–(viii).
12 CMES, 2022

4.4 Choices of Design Parameters

The design parameters for the proposed HEWMA1 and HEWMA2 charts are the smoothing
constants λ1 , λ2 and the chart width coefficient Lproposed , which have a certain effect on the
chart performance. Therefore, the different settings of the design parameters are used in comput-
ing ARL, MDRL and SDRL measures. The various combination of smoothing parameters are
chosen as (λ1 = 0.05, λ2 = 0.050001, 0.1, 0.20), (λ1 = 0.1, λ2 = 0.05, 0.100001, 0.20), (λ1 = 0.2,
λ2 = 0.05, 0.1, 0.200001), (λ1 = 0.3, λ2 = 0.05, 0.1, 0.2) to determine the values of L+
HEWMA1 and
+
LHEWMA2 , so that ARL0 = 200. The numerical results regarding the proposed HEWMA1 and
HEWMA2 charts are displayed in Tables 1–4.

5 Comparative Study
This section addresses the detailed comparative study of the proposed charts to the existing
charts. The ARL values in Tables 1–4 reveal that the HEWMA2 chart outperforms the HEWMA1
chart; therefore, the HEWMA2 chart is recommended to compare with the existing charts for
better detection performance. Thus the HEWMA2 charts is compared against the existing CH [16],
CEWMA [19], HEWMA [23], AEWMA [45], AIBEWMA1, and AIBEWMA2 [22] charts. Table 5
presents the ARL values for comparison, while Table 6 contains the overall performance values.
The following Subsections offer further details about the comparisons.

5.1 Proposed vs. CH Chart

The proposed HEWMA2 chart achieves better performance against the CH chart. For exam-
ple, at ARL0 = 200, with λ1 = 0.1, 0.2, λ2 = 0.05 and δ = 1.1, the proposed HEWMA2 charts
provide the ARL1 values 25.40 and 26.91, respectively, whereas the CH chart produces the
ARL1 = 44.26, 46.63, respectively (see Table 5 & Fig. 1). Similarly, the proposed HEWMA2 chart
indicates improved overall performance against the CH chart. As for λ1 = 0.1 and λ2 = 0.05 the
EQL, PCI, and RARL values of the proposed HEWMA2 charts are 19.5141, 1.0000, and 1.0000,
which are less than the EQL, PCI, and RARL values of the CH chart; 27.7505, 1.4221, and
1.8851 (see Table 6).

5.2 Proposed vs. CEWMA Chart

The proposed HEWMA2 chart shows lower ARL1 values when it is compared to the
CEWMA chart. For instance, assuming ARL0 = 200, with λ1 = 0.1, 0.2, λ2 = 0.05 and δ = 1.1
the proposed HEWMA2 chart has ARL1 values of 25.40 and 26.91, respectively, whereas the
CEWMA chart owns the ARL1 values of 31.80 and 37.80 (see Table 5 & Fig. 1). Similarly, in
terms of overall performance (see Table 6), the proposed HEWMA2 charts attained smaller the
EQL, PCI, and RARL values, i.e., 20.1797, 1.0000, and 1.0000 against the CEWMA chart the
EQL, PCI, and RARL values, i.e., 23.5747, 1.1682, and 1.2818, respectively, when λ1 = 0.2, and
λ2 = 0.05.

5.3 Proposed vs. HEWMA Chart

The proposed HEWMA2 chart achieves superior performance over the HEWMA chart. For
instance, at ARL0 = 200, λ1 = 0.1, λ2 = 0.05 and δ = 1.2, 1.3, 1.4, 1.5 the proposed HEWMA2 chart
gives the ARL1 values 9.94, 5.73, 3.87, 2.87, whereas the HEWMA chart yields the ARL1 = 10.11,
5.75, 3.92, 2.96 (see Table 5 & Fig. 2). Similarly, the proposed HEWMA2 chart indicates improved
overall performance against the HEWMA chart. For example, the proposed charts deliver the
EQL, PCI, and RARL values as 19.5141, 1.0000, 1.0000, respectively; however, the HEWMA
CMES, 2022 13

chart provides the EQL, PCI, and RARL values as 20.6981, 1.0607, and 1.1126, respectively (see
Table 6).

5.4 Proposed vs. AEWMA Chart

Haq [45] developed the adaptive EWMA (AEWMA) chart for monitoring the process vari-
ance. The proposed HEWMA2 chart is compared to the AEWMA chart at ARL0 = 200, and the
results indicate that the proposed HEWMA1 and HEWMA2 chart has better detection ability
against the AEWMA chart for the small shift, i.e., 1 < δ ≤ 1.2. For instance, with ARL0 = 200,
λ1 = 0.1, λ2 = 0.05 and δ = 1.1, 1.2, the proposed HEWMA-2 charts deliver the ARL1 values as
25.40, 9.94, while the AEWMA chart has the ARL1 values equal to 26.04, 10.35 (see Table 5
& Fig. 2). Likewise, when λ1 = 0.2 and λ2 = 0.05, the overall performance of the proposed
HEWMA2 chart is superior to the AEWMA chart as the proposed chart has a smaller EQL =
20.1797 than the AEWMA chart EQL = 20.3854 (see Table 6).

5.5 Proposed vs. AIBEWMA1 and AIBEWMA2 Charts

Haq [22] proposed the AIBEWMA1 and AIBEWMA2 charts for monitoring the process
variance. The comparison of the HEWMA2 chart against the AIBEWMA1 and AIBEWMA2
charts demonstrates that the proposed HEWMA2 chart is more efficient than the AIBEWMA1
and AIBEWMA2 charts. For example, with chart properties, i.e., ARL0 = 200, λ1 = 0.1, λ2 = 0.05,
ρ = 0.5 and δ = 1.1 the ARL1 values for the proposed HEWMA2 charts is observed as 25.40,
while the ARL1 values for the AIBEWMA1 and AIBEWMA2 charts are reported as 30.11 and
30.79 (see Table 5 & Fig. 3). Likewise, the proposed charts’ EQL, PCI, and RARL values also
show the edge in the overall detection ability of the HEWMA2 chart over the AIBEWMA1
and AIBEWMA2 charts. As the EQL, PCI, and RARL values for the proposed HEWMA2
charts are 19.5141, 1.0000, and 1.0000, respectively, where EQL, PCI, and RARL values for the
AIBEWMA1 and AIBEWMA2 charts are 20.8605, 1.0690 and 1.1185, and 21.4587, 1.0996 and
1.1919, respectively (see Table 6).

6 Important Points of the Study

A few important points related to the HEWMA1 and HEWMA2 charts can be listed as
(i) The HEWMA statistics undoubtedly boost the efficiency of the proposed HEWMA1 and
HEWMA2 charts.
(ii) The proposed HEWMA2 chart has better detection performance than the proposed
HEWMA1 chart (see Tables 1–4).
(iii) At different parametric settings, the ARL1 values for the proposed HEWMA1 and
HEWMA2 charts are less than the ARL1 values of the CH, CEWMA, HEWMA,
AEWMA, AIBEWMA1, and AIBEWMA2 charts (see Section 5).
(iv) The overall performance reveals the dominance of HEWMA1 and HEWMA2 charts over
the CH, CEWMA, HEWMA, AEWMA, AIBEWMA1, and AIBEWMA2 charts (see
Section 5).
(v) The proposed HEWMA1 and HEWMA2 charts have better ARL1 performance for smaller
λ1 and λ2 (see Tables 1–4).
(vi) The control limit coefficient L for the suggested HEWMA1 and HEWMA2 charts increases
as λ1 and λ2 increases.
 14

Table 1: Run-length properties of proposed HEWMA1 chart when λ1 is small and ARL0 = 200
λ1 = 0.05 λ1 = 0.10
λ2 = 0.05 λ2 = 0.10 λ2 = 0.20 λ2 = 0.05 λ2 = 0.10 λ2 = 0.20
+ + + + + +
LHEWMA1 = 1.211 LHEWMA1 = 1.366 LHEWMA1 = 1.506 LHEWMA1 = 1.365 LHEWMA1 = 1.538 LHEWMA1 = 1.691
δ ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL
1.0 200.24 112 250.70 200.67 124 233.27 200.57 129 222.93 200.29 124 232.93 200.08 132 218.11 200.30 135 210.17
1.1 25.20 14 30.05 27.60 18 30.36 28.96 20 30.20 27.52 18 30.33 30.60 21 31.43 32.71 23 32.28
1.2 10.00 5 11.70 11.17 7 11.81 11.91 8 11.65 11.15 7 11.78 12.62 9 12.11 13.54 10 12.27
1.3 5.83 3 6.49 6.47 4 6.71 6.88 5 6.68 6.48 4 6.74 7.29 5 6.95 7.81 6 6.93
1.4 3.94 2 4.32 4.40 3 4.47 4.72 3 4.41 4.38 3 4.45 4.97 3 4.61 5.32 4 4.57
1.5 2.94 2 3.05 3.28 2 3.15 3.54 2 3.21 3.29 2 3.15 3.70 2 3.35 4.04 3 3.40
1.6 2.39 1 2.31 2.65 2 2.48 2.83 2 2.48 2.65 2 2.48 2.97 2 2.60 3.21 2 2.60
1.7 2.03 1 1.81 2.22 1 1.94 2.38 2 2.00 2.21 1 1.94 2.50 2 2.12 2.68 2 2.10
1.8 1.78 1 1.52 1.94 1 1.60 2.09 1 1.67 1.94 1 1.61 2.14 1 1.72 2.32 2 1.79
1.9 1.61 1 1.26 1.75 1 1.35 1.84 1 1.38 1.74 1 1.34 1.91 1 1.45 2.05 1 1.54
2.0 1.48 1 1.06 1.60 1 1.18 1.69 1 1.20 1.61 1 1.17 1.74 1 1.26 1.85 1 1.30
CMES, 2022
 CMES, 2022

Table 2: Run-length properties of proposed HEWMA1 chart when λ1 is moderate and ARL0 = 200
λ1 = 0.20 λ1 = 0.30
λ2 = 0.05 λ2 = 0.10 λ2 = 0.20 λ2 = 0.05 λ2 = 0.10 λ2 = 0.20
+ + + + + +
LHEWMA1 = 1.506 LHEWMA1 = 1.692 LHEWMA1 = 1.849 LHEWMA1 = 1.579 LHEWMA1 = 1.773 LHEWMA1 = 1.928
δ ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL
1.0 200.57 129 222.93 200.89 136 210.79 200.64 138 204.52 200.26 130 220.98 200.64 136 207.68 200.52 140 201.14
1.1 28.96 20 30.20 32.68 23 32.22 35.94 26 34.30 29.33 20 29.99 33.54 24 32.49 37.37 26 35.74
1.2 11.91 8 11.65 13.54 10 12.32 14.80 11 12.95 12.09 9 11.54 13.91 11 12.38 15.30 12 13.29
1.3 6.88 5 6.68 7.84 6 6.93 8.48 7 7.05 7.04 5 6.59 8.03 6 6.85 8.69 7 7.06
1.4 4.72 3 4.41 5.31 4 4.58 5.80 5 4.67 4.81 3 4.32 5.45 4 4.51 5.98 5 4.65
1.5 3.54 2 3.21 4.05 3 3.38 4.36 3 3.38 3.63 3 3.15 4.14 3 3.34 4.49 4 3.37
1.6 2.83 2 2.48 3.20 2 2.60 3.49 3 2.72 2.92 2 2.48 3.31 2 2.57 3.59 3 2.66
1.7 2.38 2 2.00 2.69 2 2.12 2.90 2 2.16 2.47 2 1.98 2.76 2 2.11 3.03 2 2.17
1.8 2.09 1 1.67 2.32 2 1.80 2.54 2 1.85 2.13 1 1.65 2.39 2 1.80 2.58 2 1.81
1.9 1.84 1 1.38 2.05 1 1.52 2.20 2 1.56 1.89 1 1.38 2.11 2 1.51 2.29 2 1.58
2.0 1.69 1 1.20 1.85 1 1.30 1.99 2 1.36 1.74 1 1.20 1.92 1 1.33 2.04 2 1.36
15
 16

Table 3: Run-length properties of proposed HEWMA2 chart when λ1 is small and ARL0 = 200
λ1 = 0.05 λ1 = 0.10
λ2 = 0.05 λ2 = 0.10 λ2 = 0.20 λ2 = 0.05 λ2 = 0.10 λ2 = 0.20
+ + + + + +
LHEWMA2 = 1.229 LHEWMA2 = 1.399 LHEWMA2 = 1.559 LHEWMA2 = 1.399 LHEWMA2 = 1.598 LHEWMA2 = 1.776
δ ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL
1.0 200.99 111 251.45 200.38 122 235.48 200.59 127 228.05 200.38 122 235.48 200.74 130 221.18 200.21 132 213.57
1.1 23.53 14 27.52 25.40 16 28.01 26.67 18 27.96 25.40 16 28.01 28.13 20 29.16 30.04 22 30.10
1.2 9.15 5 10.60 9.94 6 10.77 10.58 7 10.90 9.94 6 10.77 11.23 8 11.46 11.93 9 11.64
1.3 5.21 3 5.96 5.73 3 6.14 6.00 4 6.00 5.73 3 6.14 6.33 4 6.31 6.75 5 6.38
1.4 3.52 2 3.76 3.87 2 3.97 4.11 3 3.97 3.87 2 3.97 4.25 3 4.20 4.54 3 4.15
1.5 2.69 1 2.73 2.87 2 2.80 3.10 2 2.93 2.87 2 2.80 3.19 2 2.95 3.40 2 3.00
1.6 2.14 1 2.01 2.37 1 2.23 2.47 2 2.17 2.37 1 2.23 2.57 2 2.28 2.78 2 2.38
1.7 1.87 1 1.63 2.00 1 1.69 2.10 1 1.72 2.00 1 1.69 2.19 1 1.83 2.35 2 1.93
1.8 1.65 1 1.33 1.75 1 1.37 1.82 1 1.41 1.75 1 1.37 1.90 1 1.51 2.04 1 1.59
1.9 1.51 1 1.09 1.59 1 1.15 1.69 1 1.24 1.59 1 1.15 1.73 1 1.31 1.80 1 1.35
2.0 1.40 1 0.90 1.46 1 0.97 1.54 1 1.05 1.46 1 0.97 1.58 1 1.09 1.62 1 1.11
CMES, 2022
 CMES, 2022

Table 4: Run-length properties of proposed HEWMA2 chart when λ1 is moderate and ARL0 = 200
λ1 = 0.20 λ1 = 0.30
λ2 = 0.05 λ2 = 0.10 λ2 = 0.20 λ2 = 0.05 λ2 = 0.10 λ2 = 0.20
+ + + + + +
LHEWMA2 = 1.558 LHEWMA2 = 1.778 LHEWMA2 = 1.986 LHEWMA2 = 1.646 LHEWMA2 = 1.881 LHEWMA2 = 2.095
δ ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL ARL MDRL SDRL
1.0 200.30 127 227.88 200.88 132 214.16 200.87 134 211.01 200.19 128 224.30 200.24 133 212.17 200.42 137 206.04
1.1 26.58 18 27.89 30.26 22 30.36 33.72 24 33.02 26.91 19 27.94 31.14 22 31.11 35.28 25 34.63
1.2 10.59 7 10.90 11.85 9 11.53 13.18 10 12.32 10.75 7 10.89 12.24 9 11.63 13.58 10 12.62
1.3 6.01 4 6.01 6.74 5 6.39 7.35 6 6.62 6.09 4 5.96 6.97 5 6.47 7.60 6 6.64
1.4 4.08 3 3.97 4.56 3 4.16 5.02 4 4.29 4.19 3 3.93 4.68 3 4.12 5.24 4 4.43
1.5 3.11 2 2.93 3.40 2 3.00 3.79 3 3.26 3.16 2 2.88 3.58 3 3.03 3.91 3 3.26
1.6 2.47 2 2.16 2.81 2 2.38 2.97 2 2.45 2.53 2 2.16 2.85 2 2.39 3.04 2 2.38
1.7 2.09 1 1.73 2.34 1 1.94 2.47 2 1.94 2.15 1 1.73 2.41 2 1.93 2.57 2 1.97
1.8 1.83 1 1.41 2.03 1 1.59 2.17 2 1.64 1.87 1 1.43 2.07 1 1.61 2.24 2 1.65
1.9 1.68 1 1.23 1.80 1 1.33 1.93 1 1.40 1.70 1 1.23 1.81 1 1.28 1.98 1 1.40
2.0 1.53 1 1.04 1.61 1 1.11 1.74 1 1.20 1.58 1 1.07 1.68 1 1.14 1.80 1 1.24
17
18 CMES, 2022

Table 5: ARL values for the proposed HEWMA1 and HEWMA2 vs. existing charts when λ2 =
0.05 at ARL0 = 200
δ
Chart 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
HEWMA1 λ1 = 0.1 200.29 27.52 11.15 6.48 4.38 3.29 2.65 2.21 1.94 1.74 1.61
λ1 = 0.2 200.26 29.33 12.09 7.04 4.81 3.63 2.92 2.47 2.13 1.89 1.74
HEWMA2 λ1 = 0.1 200.38 25.40 9.94 5.73 3.87 2.87 2.37 2.00 1.75 1.59 1.46
λ1 = 0.2 200.19 26.91 10.75 6.09 4.19 3.16 2.53 2.15 1.87 1.70 1.58
HEWMA λ1 = 0.1 200.04 25.50 10.11 5.75 3.92 2.96 2.40 2.02 1.79 1.62 1.49
λ1 = 0.2 201.11 26.82 10.73 6.19 4.19 3.18 2.54 2.14 1.88 1.69 1.55
AEWMA λ1 = 0.1 200.76 26.04 10.35 5.71 3.78 2.83 2.29 1.97 1.75 1.58 1.47
λ1 = 0.2 200.23 29.00 11.13 6.14 4.09 3.07 2.47 2.09 1.84 1.65 1.52
AIBEWMA1 λ1 = 0.1 200.06 30.11 11.52 6.42 4.31 3.25 2.63 2.21 1.94 1.75 1.61
λ1 = 0.2 200.09 37.00 14.00 7.61 5.00 3.70 2.92 2.44 2.13 1.89 1.73
AIBEWMA2 λ1 = 0.1 199.93 30.79 12.17 6.90 4.68 3.52 2.83 2.40 2.09 1.86 1.71
λ1 = 0.2 200.95 36.03 13.86 7.74 5.19 3.88 3.09 2.59 2.24 1.99 1.82
CH λ1 = 0.1 200.02 44.26 18.23 10.56 7.35 5.68 4.68 4.02 3.56 3.22 2.95
λ1 = 0.2 200.64 46.63 18.79 10.54 7.16 5.41 4.38 3.73 3.27 2.92 2.67
CEWMA λ1 = 0.1 200.82 31.80 12.53 7.15 4.90 3.61 2.94 2.48 2.16 1.93 1.76
λ1 = 0.2 200.24 37.80 14.74 8.18 5.50 4.08 3.24 2.70 2.33 2.07 1.88

Table 6: Overall performance measures for the proposed HEWMA1 and HEWMA2 vs. existing
charts
λ1 = 0.1, λ2 = 0.05 λ1 = 0.2, λ2 = 0.05
Charts EQL PCI RARL EQL PCI RARL
HEWMA1 20.5397 1.0526 1.1091 21.4339 1.0622 1.1272
HEWMA2 19.5141 1.0000 1.0000 20.1797 1.0000 1.0000
HEWMA 19.6104 1.0049 1.0143 20.2266 1.0023 1.0009
AEWMA 19.6088 1.0049 1.0075 20.3854 1.0102 1.0025
AIBEWMA1 20.8605 1.0690 1.1185 22.7672 1.1282 1.1878
AIBEWMA2 21.4587 1.0996 1.1919 22.9490 1.1372 1.2238
CH 27.7505 1.4221 1.8851 27.6287 1.3691 1.6911
CEWMA 21.8921 1.1219 1.2317 23.5747 1.1682 1.2818

7 Real-life Application of the Proposed Charts

This subsection explains the application of the proposed HEWMA1 and HEWMA2 charts
to real-life data. For this purpose, the real-life data are considered, representing the inside diam-
eter of the cylinder bores in an engine block. These real-life data are used by [46,47] in their
studies. The data comprise 30 samples, each size n = 5 given in Table 7. In order to imple-
ment the proposed HEWMA1 and HEWMA2 control along with CH and CEWMA charts,
following [23,48], an upward shift of size δ = 1.25 is introduced artificially after sample number
16 [49,50]. At ARL0 = 200, the smoothing parameter values λ1 = 0.1, λ2 = 0.05 are used, which
CMES, 2022 19

provides the width of the the HEWMA1, and HEWMA2, CH and CEWMA charts, respec-
tively, given as L+ + + +
HEWMA1 = 1.365, LHEWMA2 = 1.399, LCH = 1.303 and LCEWMA = 2.198. Using
the aforementioned parameters, the charting statistics for the HEWMA1, and HEWMA2, CH
and CEWMA charts and their corresponding upper control limits are computed. The chart-
ing statistics for HEWMA1, and HEWMA2, CH and CEWMA charts are given in Table 7,
while their corresponding upper control limits are given as 0.1528, 0.1314, 0.2547, and 0.4876,
respectively. Figs. 4–7 display the charting statistics of the CH, CEWMA, HEWMA1, and
HEWMA2 charts against sample number. The results show that the proposed HEWMA1 chart
outperforms the CH chart as the proposed HEWMA1 chart trigger the first OOC point after
sample number 25, while the CH chart detects OOC point after sample number 28. Overall,
the proposed HEWMA1 chart declares 7 OOC points, while the CH chart detects 2 OOC
signals. Similarly, the HEWMA2 chart gains better detection ability relative to the CEWMA
chart, as as the proposed HEWMA2 chart identifies the first OOC signal at sample number 26,
while the CH chart diagnoses OOC signal at sample number 29. This indicates that the proposed
HEWMA1 is more efficient than the CH chart, and the HEWMA2 chart achieves better detection
ability than the CHWMA chart.

Table 7: Real-life data along with charting statistics for CH, CEWMA, HEWMA1, and
HEWMA2 charts
t X1t X2t X3t X4t X5t St2 Wt Q+
t Ut+ Tt Zt Vt
1 205 202 204 207 205 3.3 –0.0835 0 0 –0.9194 –0.0919 –0.0038
2 202 196 201 198 202 7.2 0.2553 0.0255 0.0013 –0.0109 –0.0838 –0.0058
3 201 202 199 197 196 6.5 0.2109 0.0441 0.0034 –0.1503 –0.0905 –0.0071
4 205 203 196 201 197 14.8 0.5682 0.0965 0.0081 1.1324 0.0318 –0.0010
5 199 196 201 200 195 6.7 0.2240 0.1092 0.0131 –0.1096 0.0177 0.0049
6 202 202 198 203 202 3.8 –0.0223 0.0961 0.0173 –0.7815 –0.0622 0.0068
7 197 196 196 200 204 11.8 0.4698 0.1335 0.0231 0.7444 0.0184 0.0136
8 199 200 204 196 202 9.2 0.3617 0.1563 0.0297 0.3474 0.0513 0.0225
9 202 196 204 195 197 15.7 0.5938 0.2000 0.0383 1.2373 0.1699 0.0376
10 205 204 202 208 205 4.7 0.0700 0.1870 0.0457 –0.5518 0.0977 0.0486
11 200 201 199 200 201 0.7 –0.7570 0.0926 0.0480 –1.8048 –0.0925 0.0491
12 205 196 201 197 198 13.3 0.5218 0.1356 0.0524 0.9463 0.0114 0.0554
13 202 199 200 198 200 2.2 –0.2596 0.0960 0.0546 –1.2545 –0.1152 0.0549
14 200 200 201 205 201 4.3 0.0314 0.0896 0.0563 –0.6512 –0.1688 0.0519
15 202 202 204 198 203 5.2 0.1139 0.0920 0.0581 –0.4333 –0.1953 0.0480
16 201 198 204 201 201 4.5 0.0512 0.0879 0.0596 –0.6010 –0.2358 0.0423
17 200 204 198 199 199 8.6 0.3321 0.1123 0.0623 0.2444 –0.1878 0.0398
18 203 200 204 199 200 7.3 0.2639 0.1275 0.0655 0.0168 –0.1673 0.0388
19 196 203 197 201 194 21.4 0.7285 0.1876 0.0716 1.8125 0.0306 0.0484
20 197 199 203 200 196 11.7 0.4668 0.2155 0.0788 0.7329 0.1009 0.0615
21 201 197 196 199 197 6.3 0.1938 0.2133 0.0855 –0.2021 0.0706 0.0725
22 204 196 201 199 197 16.1 0.6046 0.2225 0.0939 1.2819 0.1917 0.0896

(Continued)
20 CMES, 2022

Table 7 (continued)
t X1t X2t X3t X4t X5t St2 Wt Q+
t Ut+ Tt Zt Vt
23 206 206 199 200 203 16.7 0.6211 0.2293 0.1037 1.3508 0.2076 0.1120
24 204 203 199 199 197 13.8 0.5362 0.2340 0.1142 1.0037 0.2372 0.1371
25 199 201 201 194 200 13.3 0.5212 0.2347 0.1252 0.9439 0.3339 0.1641
26 201 196 197 204 200 16.1 0.6046 0.2417 0.1370 1.2819 0.3787 0.1943
27 203 197 199 197 201 10.6 0.4243 0.2480 0.1486 0.5732 0.4541 0.2234
28 203 197 199 197 201 10.6 0.4243 0.2494 0.1598 0.5732 0.4990 0.2515
29 197 194 199 200 199 8.9 0.3476 0.2550 0.1704 0.2981 0.5059 0.2770
30 200 201 200 197 200 3.6 –0.0465 0.2693 0.1783 –0.8374 0.3716 0.2941
31 199 199 201 201 201 1.9 –0.3291 0.2334 0.1826 –1.3633 0.1981 0.3010
32 200 204 197 197 199 13.0 0.5108 0.2482 0.1879 0.9032 0.2686 0.3114

Figure 1: Comparison of proposed HEWMA1 and HEWMA2 charts with CH and CEWMA
charts at (λ1 , λ2 ) = 0.1 and ARL0 = 200

Figure 2: Comparison of proposed HEWMA1 and HEWMA2 charts with HEWMA and
AEWMA charts at (λ1 , λ2 ) = 0.1 and ARL0 = 200
CMES, 2022 21

Figure 3: Comparison of proposed HEWMA1 and HEWMA2 charts with AIBEWMA1 and
AIBEWMA2 charts at (λ1 , λ2 ) = 0.1 and ARL0 = 200

Figure 4: Real-life application of CH chart using (λ1 , λ2 ) = (0.1, 0.05) and ARL0 = 200

Figure 5: Real-life application of proposed HEWMA1 chart using (λ1 , λ2 ) = (0.1, 0.05) and
ARL0 = 200
22 CMES, 2022

Figure 6: Real-life application of CEWMA chart using (λ1 , λ2 ) = (0.1, 0.05) and ARL0 = 200

Figure 7: Real-life application of proposed HEWMA2 chart using (λ1 , λ2 ) = (0.1, 0.05) and
ARL0 = 200

8 Concluding Remarks
This paper proposes two new hybrid EWMA charts to monitor the shifts in the process
variance. The proposed charts are called HEWMA1 and HEWMA2 charts. The HEWMA1
chart is designed using the CH statistic as the input for the HEWMA1 statistic, while in the
same lines, CEWMA statistic is used as the input for the HEWMA2 statistic to construct the
proposed HEWMA2 chart. In order to evaluate the performance of the proposed HEWMA1 and
HEWMA2 charts, the extensive Monte Carlo simulation approach is used to approximate the run
length properties, including the average run length and standard deviation run length. Similarly,
to assess the overall performances of the proposed HEWMA1 and HEWMA2 charts, the extra
quadratic loss, relative average run length, and performance comparison index are computed. The
proposed HEWMA1 and HEWMA2 charts are compared to existing CH, CEWMA, HEWMA,
AEWMA, HHW1, HHW2, AIBEWMA1, and AIBEWMA2 charts, and the comparison indicates
that the proposed HEWMA1 and HEWMA2 charts outperform the existing charts. In the end,
CMES, 2022 23

real-life data are analyzed to enhance the efficiency of the proposed HEWMA1 and HEWMA2
charts.

Funding Statement: 2019 Shanxi Province Soft Science Research Program Project “Research on
Sustainable Development Capacity and Classification Construction of Shanxi Development Zone”
(Project No. 2019041005-2).
Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding
the present study.

References
1. Ajadi, J. O., Riaz, M., Al-Ghamdi, K. (2016). On increasing the sensitivity of mixed EWMA-
CUSUM control charts for location parameter. Journal of Applied Statistics, 43(7), 1262–1278. DOI
10.1080/02664763.2015.1094453.
2. Montgomery, D. C. (2012). Introduction to statistical quality control. New York: John Willey & Sons.
3. Roberts, S. (1959). Control chart tests based on geometric moving averages. Technometrics, 1(3), 239–250.
DOI 10.1080/00401706.1959.10489860.
4. Hunter, J. S. (1986). The exponentially weighted moving average. Journal of Quality Technology, 18(4),
203–210. DOI 10.1080/00224065.1986.11979014.
5. Crowder, S. V. (1987). A simple method for studying run-length distributions of exponentially weighted
moving average charts. Technometrics, 29(4), 401–407. DOI 10.2307/1269450.
6. Crowder, S. V. (1989). Design of exponentially weighted moving average schemes. Journal of Quality
Technology, 21(3), 155–162. DOI 10.1080/00224065.1989.11979164.
7. Lucas, J. M., Saccucci, M. S. (1990). Exponentially weighted moving average control schemes—Properties
and enhancements. Technometrics, 32(1), 1–12. DOI 10.1080/00401706.1990.10484583.
8. Haq, A. (2013). A new hybrid exponentially weighted moving average control chart for monitoring process
mean. Quality and Reliability Engineering International, 29(7), 1015–1025. DOI 10.1002/qre.1453.
9. Abbas, N., Riaz, M., Does, R. J. (2014). Memory-type control charts for monitoring the process dispersion.
Quality and Reliability Engineering International, 30(5), 623–632. DOI 10.1002/qre.1514.
10. Ali, R., Haq, A. (2018). A mixed GWMA-CUSUM control chart for monitoring the process mean. Com-
munications in Statistics—Theory and Methods, 47(15), 3779–3801. DOI 10.1080/03610926.2017.1361994.
11. Tang, A., Sun, J., Hu, X., Castagliola, P. (2019). A new nonparametric adaptive EWMA control
chart with exact run length properties. Computers & Industrial Engineering, 130(3), 404–419. DOI
10.1016/j.cie.2019.02.045.
12. Haq, A. (2019). A new nonparametric synthetic EWMA control chart for monitoring pro-
cess mean. Communications in Statistics—Simulation and Computation, 48(6), 1665–1676. DOI
10.1080/03610918.2017.1422750.
13. Rasheed, Z., Zhang, H., Anwar, S. M., Zaman, B. (2021). Homogeneously mixed memory charts with
application in the substrate production process. Mathematical Problems in Engineering, 2021, 1–15. DOI
10.1155/2021/2582210.
14. Rasheed, Z., Zhang, H., Arslan, M., Zaman, B., Anwar, S. M. et al. (2021). An efficient robust nonpara-
metric triple EWMA Wilcoxon signed-rank control chart for process location. Mathematical Problems in
Engineering, 2021(4), 1–28. DOI 10.1155/2021/2570198.
15. Domangue, R., Patch, S. C. (1991). Some omnibus exponentially weighted moving average statistical process
monitoring schemes. Technometrics, 33(3), 299–313. DOI 10.1080/00401706.1991.10484836.
16. Crowder, S. V., Hamilton, M. D. (1992). An EWMA for monitoring a process standard deviation. Journal
of Quality Technology, 24(1), 12–21. DOI 10.1080/00224065.1992.11979369.
17. Shu, L., Jiang, W. (2008). A new EWMA chart for monitoring process dispersion. Journal of Quality
Technology, 40(3), 319–331. DOI 10.1080/00224065.2008.11917737.
24 CMES, 2022

18. Huwang, L., Huang, C. J., Wang, Y. H. T. (2010). New EWMA control charts for monitoring process
dispersion. Computational Statistics & Data Analysis, 54(10), 2328–2342. DOI 10.1016/j.csda.2010.03.011.
19. Castagliola, P. (2005). A new S2-EWMA control chart for monitoring the process variance. Quality and
Reliability Engineering International, 21(8), 781–794. DOI 10.1002/(ISSN)1099-1638.
20. Chang, T. C., Gan, F. F. (1994). Optimal designs of one-sided EWMA charts for monitoring a process vari-
ance. Journal of Statistical Computation and Simulation, 49(1–2), 33–48. DOI 10.1080/00949659408811559.
21. Razmy, A. M., Peiris, T. S. G. (2015). Design of exponentially weighted moving average chart for monitoring
standardized process variance. International Journal of Engineering & Technology, 13(5), 74–78.
22. Haq, A. (2017). New EWMA control charts for monitoring process dispersion using auxiliary information.
Quality and Reliability Engineering International, 33(8), 2597–2614. DOI 10.1002/qre.2220.
23. Ali, R., Haq, A. (2017). New memory-type dispersion control charts. Quality and Reliability Engineering
International, 33(8), 2131–2149. DOI 10.1002/qre.2174.
24. Saghir, A., Aslam, M., Faraz, A., Ahmad, L., Heuchenne, C. (2020). Monitoring process variation using
modified EWMA. Quality and Reliability Engineering International, 36(1), 328–339. DOI 10.1002/qre.2576.
25. Zaman, B., Lee, M. H., Riaz, M., Abujiya, M. R. (2019). An adaptive approach to EWMA dispersion chart
using Huber and Tukey functions. Quality and Reliability Engineering International, 35(6), 1542–1581. DOI
10.1002/qre.2460.
26. Riaz, M., Abbasi, S. A., Abid, M., Hamzat, A. K. (2020). A new HWMA dispersion control chart with an
application to wind farm data. Mathematics, 8(12), 2136. DOI 10.3390/math8122136.
27. Chatterjee, K., Koukouvinos, C., Lappa, A. (2020). A new S2-TEWMA control chart for monitoring process
dispersion. Quality and Reliability Engineering International, 37(4), 1334–1354. DOI 10.1002/qre.2798.
28. Haq, A. (2016). A new hybrid exponentially weighted moving average control chart for monitoring
process mean: Discussion. Quality and Reliability Engineering International, 33(7), 1629–1631. DOI
10.1002/qre.2092.
29. Aslam, M., Azam, M., Khan, N., Jun, C. H. (2015). A control chart for an exponential distribution using
multiple dependent state sampling. Quality & Quantity, 49(2), 455–462. DOI 10.1007/s11135-014-0002-2.
30. Aslam, M., Khan, N., Jun, C. (2016). A hybrid exponentially weighted moving average chart for COM-
Poisson distribution. Transactions of the Institute of Measurement and Control, 40(2), 456–461. DOI
10.1177/0142331216659920.
31. Aslam, M., Azam, M., Jun, C. H. (2018). A HEWMA-CUSUM control chart for the Weibull distribution.
Communications in Statistics—Theory and Methods, 47(24), 5973–5985. DOI 10.1080/03610926.2017.
1404100.
32. Noor-ul-Amin, M., Khan, S., Sanaullah, A. (2019). HEWMA control chart using auxiliary infor-
mation. Iranian Journal of Science and Technology, Transactions A: Science, 43(3), 891–903. DOI
10.1007/s40995-018-0585-x.
33. Aslam, M., Rao, G. S., AL-Marshadi, A. H., Jun, C. H. (2019). A nonparametric HEWMA-p control chart
for variance in monitoring processes. Symmetry, 11(3), 356. DOI 10.3390/sym11030356.
34. Noor, S., Noor-ul-Amin, M., Mohsin, M., Ahmed, A. (2020). Hybrid exponentially weighted moving
average control chart using Bayesian approach. Communications in Statistics—Theory and Methods. DOI
10.1080/03610926.2020.1805765.
35. Asif, F., Khan, S., Noor-ul-Amin, M. (2020). Hybrid exponentially weighted moving average control chart
with measurement error. Iranian Journal of Science and Technology, Transactions A: Science, 44(3), 801–811.
DOI 10.1007/s40995-020-00879-3.
36. Noor-Ul-Amin, M., Shabbir, N., Khan, N., Riaz, A. (2020). Hybrid exponentially weighted moving average
control chart for mean using different ranked set sampling schemes. Kuwait Journal of Science, 47(4), 19–28.
37. Lawless, J. F. (2003). Statistical models and methods for lifetime data. New Jersey: John Wiley and Sons.
38. Quesenberry, C. P. (1995). On properties of Q charts for variables. Journal of Quality Technology, 27(3),
184–203. DOI 10.1080/00224065.1995.11979592.
39. Awais, M., Haq, A. (2018). A new cumulative sum control chart for monitoring the process mean using
varied L ranked set sampling. Journal of Industrial and Production Engineering, 35(2), 74–90. DOI
10.1080/21681015.2017.1417787.
CMES, 2022 25

40. Aslam, M., Anwar, S. M. (2020). An improved Bayesian Modified-EWMA location chart and its applica-
tions in mechanical and sport industry. PLoS One, 15(2), e0229422. DOI 10.1371/journal.pone.0229422.
41. Zafar, R. F., Abbas, N., Riaz, M., Hussain, Z. (2014). Progressive variance control charts for moni-
toring process dispersion. Communications in Statistics-Theory and Methods, 43(23), 4893–4907. DOI
10.1080/03610926.2012.717668.
42. Anwar, S. M., Aslam, M., Riaz, M., Zaman, B. (2020). On mixed memory control charts based on auxiliary
information for efficient process monitoring. Quality and Reliability Engineering International, 36(6), 1949–
1968. DOI 10.1002/qre.2667.
43. Ou, Y., Wu, Z., Tsung, F. (2012). A comparison study of effectiveness and robustness of control charts
for monitoring process mean. International Journal of Production Economics, 135(1), 479–490. DOI
10.1016/j.ijpe.2011.08.026.
44. Anwar, S. M., Aslam, M., Ahmad, S., Riaz, M. (2019). A modified-mxEWMA location chart for the
improved process monitoring using auxiliary information and its application in wood industry. Quality
Technology & Quantitative Management, 17, 561–579.
45. Haq, A. (2018). A new adaptive EWMA control chart for monitoring the process dispersion. Quality and
Reliability Engineering International, 34(5), 846–857. DOI 10.1002/qre.2294.
46. Zaman, B., Lee, M. H., Riaz, M., Abujiya, M. R. (2017). An adaptive EWMA scheme-based CUSUM accu-
mulation error for efficient monitoring of process location. Quality and Reliability Engineering International,
33(8), 2463–2482. DOI 10.1002/qre.2203.
47. Zaman, B. (2021). Efficient adaptive CUSUM control charts based on generalized likelihood ratio test to
monitor process dispersion shift. Quality and Reliability Engineering International, 37(8), 3192–3220. DOI
10.1002/qre.2903.
48. Abbas, N., Riaz, M., Does, R. J. M. M. (2013). CS-EWMA chart for monitoring process dispersion. Quality
and Reliability Engineering International, 29(5), 653–663. DOI 10.1002/qre.1414.
49. Anwar, S. M., Aslam, M., Zaman, B., Riaz, M. (2021). Mixed memory control chart based on auxiliary
information for simultaneously monitoring of process parameters: An application in glass field. Computers
& Industrial Engineering, 156(3), 107284. DOI 10.1016/j.cie.2021.107284.
50. Anwar, S. M., Aslam, M., Zaman, B., Riaz, M. (2021). An enhanced double homogeneously weighted
moving average control chart to monitor process location with application in automobile field. Quality and
Reliability Engineering International. DOI 10.1002/qre.2966.