You are on page 1of 32

Received: 14 July 2021 Revised: 15 October 2021 Accepted: 11 November 2021

DOI: 10.1002/qre.3032

RESEARCH ARTICLE

A homogeneously weighted moving average control chart


for monitoring time between events

Muhammad Aslam1 Majid Khan1,2 Syed Masroor Anwar3 Babar Zaman4

1 Department of Mathematics and

Statistics, Riphah International Abstract


University, Islamabad, Pakistan In high-quality processes, non-conforming events rarely occur, and the time
2Government Postgraduate College between events (TBE) is more likely to follow a skewed distribution, such as
Haripur, Khyber Pakhtunkhwa, Pakistan
3
gamma distribution. This paper proposes a gamma distribution-based one-sided
Department of Statistics, University of
Azad Jammu and Kashmir, homogeneously weighted moving average control chart for monitoring the TBE.
Muzaffarabad, Pakistan The proposed control chart is known as the HWMA_TBE control chart. The
4 Department of Mathematics, College of
HWMA_TBE control chart monitors the downward shifts in the process, that
Science, University of Hafr Al Batin, Hafr
is, a decrease in inter-arrival time that can lead to the process deterioration.
Al Batin, Saudi Arabia
The Monte Carlo simulation approach is used to compute specific performance
Correspondence indicators, such as average run length (ARL), standard deviation run length
Majid Khan, Department of Mathemat-
ics and Statistics, Riphah International
(SDRL), extra quadratic loss (EQL), relative average run length (RARL), and per-
University, Islamabad, Government formance comparison index (PCI). Based on these performance indicators, the
Postgraduate College Haripur, Khyber HWMA_TBE control chart is compared to some competing control charts. The
Pakhtunkhwa, Pakistan
Email: 285004majidkhan@gmail.com comparison suggests that the HWMA_TBE control chart is more efficient than
the competing control charts. Moreover, this paper also investigates the robust-
ness of the HWMA_TBE control chart, which indicates that the HWMA_TBE
control chart is more robust than the competing control charts. In the end, two
examples are provided, one with boring machine failure data and the other with
breast cancer survival data, to evaluate the implementation of the HWMA_TBE
control chart for practical situations.

KEYWORDS
average run length, gamma distribution, monte carlo simulations, robustness, time between
events

1 INTRODUCTION

Statistical process control (SPC) kit is a collection of tools used to improve the process quality and productivity. The control
chart is one of the most powerful tools in the SPC kit, used to detect the potential shifts in process parameters(s). Shewhart1
was the first to introduce the control chart, known as the Shewhart control chart, which is the most basic and widely used
control chart. The Shewhart 𝑋̄ control chart used to monitor the mean level of the process, while the process variability
is monitored by the Shewhart 𝑅 control chart. Similarly, the Shewhart 𝑐 and Shewhart 𝑢 control charts are known as the
Shewhart attribute control charts, used to monitor the number of non-conformities or defect rates. The Shewhart attribute
control charts are user-friendly, but they are usually less sensitive in monitoring the defect rates for high-quality processes.
Instead of monitoring the defect rate, an alternative approach uses the control charts to monitor the inter-arrival time of

Qual Reliab Engng Int. 2021;1–32. wileyonlinelibrary.com/journal/qre © 2021 John Wiley & Sons Ltd. 1
2 ASLAM et al.

non-conforming products followed by the exponential distribution.2 The control charts that monitor the inter-arrival time
of non-conforming products are known as the time between events (TBE) control charts. In high-quality processes, these
control charts are usually preferred over the Shewhart attribute control charts.
In literature, numerous TBE control charts are available for monitoring the shifts in high-quality processes. For exam-
ple, Lucas3 designed the cumulative sum (CUSUM) control chart for exponential data, also known as the exponential
CUSUM control chart. Further, Vardeman and Ray4 used integral equations to compute the exponential CUSUM control
chart’s average run length (ARL). Additionally, Gan5 presented the optimum design structure for the exponential CUSUM
control chart. Similarly, Gan2 proposed the exponentially weighted moving average (EWMA) control chart for exponential
data, called the exponential EWMA control chart. Likewise, Chan, et al.6 designed the cumulative quantity control (𝐶𝑄𝐶)
control chart, also known as 𝑡-control chart, that monitors the TBE in a high-quality processes. Equally, Xie, et al.7 devel-
oped the Shewhart-type control chart, denoted as 𝑡𝑟 -control chart, to monitor the time between 𝑟th failures in the gamma
distribution. Correspondingly, Zhang, et al.8 studied the gamma distribution-based Shewhart-type control chart in order
to monitor the TBE data. Santiago and Smith9 presented the design structure of the run rules-based 𝑡-control chart, which
is based on the transformed exponential distribution. Rakitzis10 investigated the two-sided ARL-unbiased Shewhart TBE
control chart that used the run rules. Chakraborty, et al.11 suggested the lower-sided generally weighted moving average
(GWMA) control chart, symbolized as GWMA_TBE control chart, to detects the decreasing shifts in the process when the
TBE observations follow the gamma distribution. Similarly, Alevizakos, et al.12 introduced the one-sided double GWMA
control chart, named DGWMA_TBE control chart, to monitor the TBE based on the gamma distribution. Likewise, Ale-
vizakos and Koukouvinos13 designed the gamma distribution-based one-sided double EWMA_TBE control chart, denoted
as DEWMA_TBE control chart, and they showed that the DEWMA_TBE control chart has better performance than the
EWMA TBE (denoted as EWMA_TBE) and Shewhart control charts. Equally, Alevizakos, et al.14 recommended one- and
two-sided triple EWMA control charts, labeled as TEWMA_TBE control charts, and demonstrated that the TEWMA_TBE
control charts have uniformly better performance than the DEWMA_TBE and EWMA_TBE control charts. For further
details about the TBE control charts, the readers are referred to the works of Liu, et al.,15 Zhang, et al.,8 Shafae, et al.,16
Yang, et al.,17 Qu, et al.,18 Kumar and Baranwal,19 Alevizakos and Koukouvinos20,21 and Hu, et al.22
Control charts with time-varying control limits, such as EWMA control charts, are generally preferred over Shewhart
control charts. The EWMA control charts used both current and previous observations of the process and, therefore, they
are more efficient in detecting small-to-moderate shifts in the process parameter(s). Hunter23 identified that the EWMA
control chart assigned a larger weight to the current observation and smaller to the previous observations. Abbas24 intro-
duced the concept of a homogeneous control chart and proposed the homogeneously weighted moving average (HWMA)
control chart. The HWMA control chart assigns a particular weight to the current observations and distributes the rest
of the weights homogeneously to the previous observations. He suggested that assigning weights in this manner would
undoubtedly enhance the detection ability of the HWMA control chart. For additional details about the homogeneous
control chart, the readers are referred to the works of Adegoke, et al.,25 Riaz, et al.,26 Riaz, et al.,27 and Alevizakos, et al.,28
and references are therein.
In practice, it is of utmost importance to detect any process deterioration as early as possible.29,30 In monitoring TBE,
the process deterioration may results from a rise in the failure rate. Consequently, a small decrease in process performance
should be early detected in a high-quality process . In this regard, the GWMA_TBE, DGWMA_TBE, and DEWMA_TBE
control charts are specifically designed to monitor the downward shifts in the process parameters. However, the high-
quality processes may require further improvement in the performances of the control charts. Moreover, as previously
stated, the HWMA control charts are an important innovation in SPC tools and have superior performance over the com-
peting control charts. It is well known that there are no HWMA control charts available to monitor the TBE, so there
is clear room to design HWMA control charts that enhance the performance of control charts for the high-quality pro-
cesses. Therefore, this paper’s primary and initial goal is to present the design structure of a gamma distribution-based
one-sided HWMA control chart with time-varying limits that efficiently monitor the TBE and effectively detect down-
ward shifts in the process. The proposed control chart is labeled as the HWMA_TBE control chart. It is worth declaring
that the process improvements, that is, a decrease in failure rate, are also essential to monitor. However, as the process
improvements occur after corrective action is taken and the time of this change is known; thus the control chart is not
required to detect them.12,31 The Monte Carlo simulations study is performed, and the numerical results for the different
performance indicators are computed. The performance of the HWMA_TBE control chart is compared to the competing
TEWMA_TBE, DEWMA_TBE, EWMA_TBE, DGWMA_TBE, and GWMA_TBE control charts. The comparison reveals
that the HWMA_TBE control chart is more efficient than the competing TBE control charts in distinguishing small-to-
moderate downward shifts in the process. Finally, two real-life examples, one with boring machine failure data and the
ASLAM et al. 3

other with breast cancer survival data, are presented to evaluate the use of the HWMA_TBE control chart in practical
scenarios.
The rest of the paper is outlined as follows: Section 2 describes the methodologies and construction of some TBE control
charts based on the gamma distribution. Similarly, Section 3 discusses the design and implementation of the HWMA_TBE
control chart. Likewise, the Phase-II monitoring of the HWMA_TBE control chart is explained in Section 4. In addition,
Section 5 offers the performance comparison of the HWMA_TBE control chart against some existing TBE control charts.
Moreover, the robustness study of the HWMA_TBE control chart, where the TBE observations are assumed to follow
the Weibull and log-normal distributions, is addressed in Section 6. Furthermore, Section 7 presents the two real-life
applications for the implementation of the HWMA_TBE control chart practically. Finally, a summary, conclusions, and
recommendations are given in Section 8.

2 CONTROL CHARTS BASED ON GAMMA DISTRIBUTION

This section sets out the methodologies and construction procedures for some control charts that monitor the TBE and
detect the downward shifts in the process when the TBE observations follow the gamma distribution. Subsection 2.1-2.4
presents the formulation of the EWMA_TBE, DEWMA_TBE, TEWMA_TBE, DGWMA_TBE, and GWMA_TBE control
charts. In addition, the design and construction of the proposed HWMA_TBE control chart are also given in this subsec-
tion.
Let 𝑇𝑖 for 𝑖 = 1, 2, … , 𝑘 be the inter-arrival times between the successive occurrences in a homogeneous Poisson pro-
cess with rate 𝜃−1 , where 𝜃 > 0, then each 𝑇𝑖 is assumed to be independently identically distributed exponential random
variable having scale parameter θ. If 𝑋 represents the sum of the inter-arrival times until the first 𝑘 failures occur, then 𝑋
∑𝑘
follows a gamma distribution, that is, 𝑋 = 𝑖=1 𝑇𝑖 ∼ Gamma(𝑘, 𝜃). In this case, the probability density function (pdf) of
𝑋 is given as follows

𝑥𝑘−1 ( 𝑥)
𝑓 (𝑥; 𝑘, 𝜃) = exp − , 𝑥, 𝑘, 𝜃 > 0, (1)
Γ (𝑘) 𝜃𝑘 𝜃

where Γ(⋅) represents the gamma function, 𝑘 is the shape, and 𝜃 is the scale parameter of the gamma distribution. Here,
𝑘 is a positive integer, so in this case, the gamma distribution can be viewed as an Erlang distribution. If the only concern
is to monitor the time until the first failure, the gamma distribution is reduced to the exponential distribution with scale
parameter 𝜃.
iid
Let 𝑋𝑡 ∼ Gamma(𝑘, 𝜃), for 𝑡 = 1, 2, … having mean 𝑘𝜃 and variance 𝑘𝜃2 . If the process is IC, then 𝜃 = 𝜃0 and if the
process is OOC, then 𝜃 = 𝛿𝜃0 , where 𝛿 (0 < 𝛿 < 1) is the downward shift in the process, and in this case, the process is
deteriorated. As a result, corrective actions should be taken to diagnose and eliminate the process deterioration sources.

2.1 EWMA_TBE control chart

Let 𝑌𝑡 denotes the plotting statistic for the EWMA_TBE control chart, then it can be defined as follows

𝑌𝑡 = 𝜆𝑋𝑡 + (1 − 𝜆) 𝑌𝑡−1 , 𝑡 = 1, 2, … , (2)

where 𝜆 is the smoothing parameter. Although the values of 𝜆 lie between 0 and 1; however, in practice, the EWMA control
chart performs better for 0.05 ≤ 𝜆 ≤ 0.25 and 𝜆 = 0.05, 0.1, and 0.2 are its common choices.32 The initial value of 𝑌𝑡 is equal
to 𝑘𝜃0 , that is, 𝑌0 = 𝑘𝜃0 . The statistic 𝑌𝑡 can also be written as follows

𝑡−1
∑ 𝑖 𝑡
𝑌𝑡 = 𝜆 (1 − 𝜆) 𝑋𝑡−𝑖 + (1 − 𝜆) 𝑘𝜃0 . (3)
𝑖=1
4 ASLAM et al.

Equation (3) can be used to find the IC mean and variance of the statistic 𝑌𝑡 , which are respectively given as;
𝜆 2𝑡
𝐸 (𝑌𝑡 |IC) = 𝑘𝜃0 and 𝑉𝑎𝑟 (𝑌𝑡 |IC) = [1 − (1 − 𝜆) ]𝑘𝜃02 . If 𝐿𝐶𝐿𝑡 denotes the lower-sided time-varying control limit
2−𝜆
for the EWMA_TBE control chart; then it can be defined as follows

𝜆 [ 2𝑡
]
𝐿𝐶𝐿𝑡 = 𝑘𝜃0 − 𝐿 1 − (1 − 𝜆) 𝑘𝜃02 , (4)
2−𝜆

where 𝐿 is the control chart width coefficient, and it is determined so that the IC ARL(ARL0 ) is equal to its pre-specified
desired value. In order to detect the decrease in interarrival times, the lower-sided EWMA_TBE control chart is designed
and 𝑌𝑡 is plotted against the 𝐿𝐶𝐿𝑡 . Whenever 𝑌𝑡 ≤ 𝐿𝐶𝐿𝑡 , the process is considered to be OOC; otherwise, the process
remains IC.

2.2 DEWMA_TBE control chart

If 𝑍𝑡 denote the plotting statistic for the DEWMA_TBE control chart; then it can be defined by the expression given as
follows

𝑍𝑡 = 𝜆𝑌𝑡 + (1 − 𝜆) 𝑍𝑡−1 , 𝑡 = 1, 2, … . (5)

The initial values of 𝑍𝑡 to be equal to 𝑘𝜃0 , that is, 𝑍0 = 𝑘𝜃0 . Alternatively, the statistic 𝑍𝑡 can also be written as follows

𝑡
∑ 𝑡−𝑖 𝑡
𝑍𝑡 = 𝜆2 (1 + 𝑡 − 𝑖) (1 − 𝜆) 𝑋𝑖 + (1 + 𝑡𝜆) (1 − 𝜆) 𝑘𝜃0 . (6)
𝑖=1

The IC mean of the statistic 𝑍𝑡 , defined by Equation (6), is 𝐸 (𝑍𝑡 |IC) = 𝑘𝜃0 , and its variance is given as follows

⎡ 2 2 2𝑡 ( 2 ) 2𝑡+2 2 − 𝜆)2𝑡+4

⎢ 1 + (1 − 𝜆) − (1 + 𝑡) (1 − 𝜆) + 2𝑡 + 𝑡 − 1 (1 − 𝜆) − 𝑡 (1 ⎥ 2
𝑉𝑎𝑟 (𝑍𝑡 |IC) = 𝜆4 ⎢ { }3 ⎥ 𝑘𝜃0 . (7)
⎢ 1 − (1 − 𝜆)
2 ⎥
⎣ ⎦

The control limit 𝐿𝐶𝐿𝑡 for the lower-sided DEWMA_TBE control chart can be used to detect decreasing shifts in the
process, is given as follows


√ ⎡ ⎤
√ 2 2 2𝑡 2𝑡+2 2𝑡+4
√ 4 ⎢ 1 + (1 − 𝜆) − (1 + 𝑡) (1 − 𝜆) + (2𝑡2 + 𝑡 − 1) (1 − 𝜆) − 𝑡2 (1 − 𝜆) ⎥ 2

𝐿𝐶𝐿𝑡 = 𝑘𝜃0 − 𝐿√𝜆 ⎢ ⎥ 𝑘𝜃0 . (8)
{ }3
⎢ 1 − (1 − 𝜆)
2 ⎥
⎣ ⎦

In this case, the process is said to be OOC when 𝑍𝑡 falls below 𝐿𝐶𝐿𝑡 , that is, 𝑍𝑡 ≤ 𝐿𝐶𝐿𝑡 ; otherwise, the process is declared
to be IC.

2.3 TEWMA_TBE control chart

The TEWMA_TBE control chart plotting statistic is denoted by 𝑊𝑡 , which can be defined as follows

𝑊𝑡 = 𝜆𝑍𝑡 + (1 − 𝜆) 𝑊𝑡−1 , 𝑡 = 1, 2, … . (9)

The starting value of 𝑊𝑡 is equal to 𝑘𝜃0 , that is, 𝑊0 = 𝑘𝜃0 . Alternatively, the statistic 𝑊𝑡 can be written as follows

𝑡 𝑡
𝜆3 ∑ 𝑡−𝑗 (1 − 𝜆)
𝑊𝑡 = (𝑡 − 𝑗 + 1) (𝑡 − 𝑗 + 2) (1 − 𝜆) 𝑋𝑗 + [𝑡𝜆 (𝑡𝜆 + 𝜆 + 2) + 2] 𝑘𝜃0 . (10)
2 𝑗=1 2
ASLAM et al. 5

The IC mean of the statistic 𝑊𝑡 is 𝐸 (𝑊𝑡 |IC) = 𝑘𝜃0 and its variance can be defined as follows
[ ( { ) ( ) ( )}
𝑡 𝑡2 − 1 (𝑡 − 2) 𝛾𝑡−3 4𝑡 𝑡2 − 1 𝛾𝑡−2 12𝑡 (𝑡 + 1) 𝛾𝑡−1 24 (𝑡 + 1) 𝛾𝑡 24 1 − 𝛾𝑡+1
𝛾3 𝜆6
Var (𝑊𝑡 |IC ) = − − − − +
4 1−𝛾 (1 − 𝛾)
2
(1 − 𝛾)
3
(1 − 𝛾)
4
(1 − 𝛾)
5

{ ( ) ( )}
2 6
𝑡 𝑡2 − 1 𝛾𝑡−2 3𝑡 (𝑡 + 1) 𝛾𝑡−1 6 (𝑡 + 1) 𝛾𝑡 6 1 − 𝛾𝑡+1
+ 2𝛾 𝜆 − − − +
1−𝛾 (1 − 𝛾)
2
(1 − 𝛾)
3
(1 − 𝛾)
4

{ } { 𝑡+1 }]
7𝛾𝜆6 𝑡(𝑡 + 1)𝛾𝑡−1 2(𝑡 + 1)𝛾𝑡 2(1 − 𝛾𝑡+1 ) 𝑡
6 − (𝑡 + 1)𝛾 + 1 − 𝛾
+ − − + + 𝛾 𝑘𝜃02 , (11)
2 1−𝛾 (1 − 𝛾)2 (1 − 𝛾)3 1−𝛾 (1 − 𝛾)2

where 𝛾 = (1 − 𝜆)2 . In order to identify the downward shifts in the process, the control limit 𝐿𝐶𝐿𝑡 , based on 𝐸(𝑊𝑡 |IC) and
𝑉𝑎𝑟(𝑊𝑡 |IC), for the lower-sided TEWMA_TBE control chart, is given as follows

𝐿𝐶𝐿𝑡 = 𝑘𝜃0 − 𝐿 𝑉𝑎𝑟(𝑊𝑡 |IC), (12)

where 𝑉𝑎𝑟(𝑊𝑡 |IC) can be determined by Equation (11). The statistic 𝑊𝑡 is plotted against the 𝐿𝐶𝐿𝑡 and if 𝑊𝑡 ≤ 𝐿𝐶𝐿𝑡 then
the process is said to be OOC, and if 𝑊𝑡 > 𝐿𝐶𝐿𝑡 then the process is deemed to be IC.

2.4 GWMA_TBE control chart

The plotting statistic for the GWMA_TBE control chart is denoted by 𝐺𝑡 and can be defined as follows

𝑡 (
∑ )
𝑎 𝑎 𝑎
𝐺𝑡 = 𝑞(𝑖−1) − 𝑞𝑖 𝑋𝑡−𝑖+1 + 𝑞𝑡 𝑘𝜃0 , (13)
𝑖=1

where 𝑞(0 ≤ 𝑞 < 1) is the design parameter, and 𝑎(𝑎 > 0) is the adjustment parameter. The IC mean and variance of the
∑𝑡 𝑎 𝑎
statistic 𝐺𝑡 are, respectively, given as; 𝐸 (𝐺𝑡 |IC) = 𝑘𝜃0 and 𝑉𝑎𝑟 (𝐺𝑡 |IC) = 𝑄𝑡 𝑘𝜃02 , where 𝑄𝑡 = 𝑖=1 (𝑞(𝑖−1) − 𝑞𝑖 ). The
lower-sided control limit 𝐿𝐶𝐿𝑡 for the GWMA_TBE control chart is given as follows

𝐿𝐶𝐿𝑡 = 𝑘𝜃0 − 𝐿 𝑄𝑡 𝑘𝜃02 . (14)

To detect the downward shifts in the process, the statistic 𝐺𝑡 is plotted against 𝐿𝐶𝐿𝑡 and the process is said to be OOC
when 𝐺𝑡 falls below 𝐿𝐶𝐿𝑡 , that is, 𝐺𝑡 ≤ 𝐿𝐶𝐿𝑡 and the process is known to be IC if 𝐺𝑡 > 𝐿𝐶𝐿𝑡 .

2.5 DGWMA_TBE control chart

The DGWMA_TBE control chart can be defined by plotting statistic 𝐷𝑡 , can be given as follows

𝑡 (
∑ )
𝑎 𝑎 𝑎
𝐷𝑡 = 𝑞(𝑖−1) − 𝑞𝑖 𝐺𝑡−𝑖+1 + 𝑞𝑡 𝑘𝜃0 . (15)
𝑖=1

The plotting statistic 𝐷𝑡 in Equation (15) can be simplified in the form given as follows

𝑡
( 𝑡
)
∑ ∑
𝐷𝑡 = 𝑤𝑖 𝑋𝑡−𝑖+1 + 1− 𝑤𝑖 𝑘𝜃0 , (16)
𝑖=1 𝑖=1

∑𝑖 (𝑗−1)
𝑎 𝑎 𝑎 𝑎
where 𝑤𝑖 = 𝑗=1 (𝑞 − 𝑞𝑖 )(𝑞(𝑖−𝑗) − 𝑞(𝑖−𝑗+1) ) . The IC mean and variance of the DGWMA_TBE control chart
∑𝑡
statistic are given as; 𝐸 (𝐷𝑡 |IC) = 𝑘𝜃0 and 𝑉𝑎𝑟 (𝐷𝑡 |IC) = 𝑅𝑡 𝑘𝜃02 , where 𝑅𝑡 = 𝑖=1 𝑤𝑖2 . The 𝐿𝐶𝐿𝑡 for the lower-sided
6 ASLAM et al.

DGWMA_TBE control chart is given as follows



𝐿𝐶𝐿𝑡 = 𝑘𝜃0 − 𝐿 𝑅𝑡 𝑘𝜃02 . (17)

In order to monitor the decrease in inter-arrival time, the lower-sided DGWMA_TBE control chart is formulated, and
the plotting statistic 𝐷𝑡 is plotted against the 𝐿𝐶𝐿𝑡 and the process is said to be OOC if 𝐷𝑡 ≤ 𝐿𝐶𝐿𝑡 ; otherwise, the process
is said to be IC.

2.6 Proposed HWMA_TBE control chart

This subsection presents a gamma distribution-based one-sided HWMA control chart to monitor the TBE and identify
the process’s decreasing shifts. This control chart is labeled as the HWMA_TBE control chart. The HWMA_TBE control
chart can be formulated by extending the idea of Abbas24 to the TBE. Defining a statistic 𝑈𝑡 is called the HWMA_TBE
statistic, given as follows

𝑈𝑡 = 𝜆𝑋𝑡 + (1 − 𝜆) 𝑋̄ 𝑡−1 , 𝑡 = 1, 2, … , (18)


∑𝑡−1
𝑋
where 𝑋̄ 𝑡−1 = 𝑖=1 𝑖 denotes the mean of previous 𝑡 − 1 observations. The initial values 𝑈0 and 𝑋̄ 0 are equal to the target
𝑡−1
value of 𝑋, that is, 𝑈0 = 𝑋̄ 0 = 𝑘𝜃0 . The plotting statistic 𝑈𝑡 defined by Equation (18) can also be rewritten as follows
[( ) ( ) ( ) ]
1−𝜆 1−𝜆 1−𝜆
𝑈𝑡 = 𝜆𝑋𝑡 + 𝑋𝑡−1 + 𝑋𝑡−2 + ⋯ + 𝑋1 , 𝑡 = 1, 2, … . (19)
𝑡−1 𝑡−1 𝑡−1

The logic behind the statistic given in Equation (19) is to assign 𝜆 weight to the current observation of the process, while
the preceding observations are homogeneously assigned the rest of the (1 − 𝜆) weights. The IC mean and variance for the
plotting statistic 𝑈𝑡 , respectively, given as follows

𝐸 (𝑈𝑡 |IC) = 𝑘𝜃0 , (20)

⎧𝜆2 𝑘𝜃2 if 𝑡 = 1,
0

𝑉𝑎𝑟 (𝑈𝑡 |IC) = ⎨[ (1−𝜆)
2
] (21)
2
⎪ 𝜆 + 𝑘𝜃02 if 𝑡 > 1.
⎩ 𝑡−1

Equations (20) and (21) are used to construct the upper control limit (𝑈𝐶𝐿𝑡 and 𝐿𝐶𝐿𝑡 for the HWMA_TBE control chart,
defined as follows

⎧ √

⎪𝑘𝜃0 + 𝐿 𝜆2 𝑘𝜃02 if 𝑡 = 1, ⎥
⎪ √[ ⎥
𝑈𝐶𝐿𝑡 = ⎨ ]
(1−𝜆)
2 ⎥
⎪𝑘𝜃0 + 𝐿 𝜆2 + 𝑘𝜃02 if 𝑡 > 1, ⎥
⎪ 𝑡−1

⎩ ⎥
⎥ (22)
⎧ √ ⎥
⎪𝑘𝜃0 − 𝐿 𝜆2 𝑘𝜃02 if 𝑡 = 1, ⎥
⎪ √[ ⎥
𝐿𝐶𝐿𝑡 = ⎨ 2
] ⎥
⎪𝑘𝜃0 − 𝐿 𝜆2 +
(1−𝜆)
𝑘𝜃02 if 𝑡 > 1. ⎥
⎪ 𝑡−1 ⎥
⎩ ⎦

Here, 𝐿 is the control chart parameter, and it is obtained in such a way that the ARL0 is equal to its desired value. If the
𝐿𝐶𝐿𝑡 calculated is less than zero, then it is reset as; 𝐿𝐶 𝐿𝑡 = 0. The lower control limit 𝐿𝐶𝐿𝑡 for the HWMA_TBE control
ASLAM et al. 7

chart that monitors downward shifts in the process is defined as follows

⎧ √
⎪𝑘𝜃0 − 𝐿 𝜆2 𝑘𝜃02 if 𝑡 = 1,
⎪ √[
𝐿𝐶𝐿𝑡 = ⎨ 2
] (23)
(1−𝜆)
⎪𝑘𝜃0 − 𝐿 𝜆2 + 𝑘𝜃02 if 𝑡 > 1.
⎪ 𝑡−1

The statistic 𝑈𝑡 is plotted along 𝐿𝐶𝐿𝑡 , and the process is said to be OOC if 𝑈𝑡 ≤ 𝐿𝐶𝐿𝑡 ; otherwise, the process is known
to be IC.

3 DESIGN AND IMPLEMENTATION OF HWMA_TBE CONTROL CHART

In this section, the run length distribution of the HWMA_TBE control chart is provided, where it is assumed that the scale
parameter of the process is known (referred to as case K). Additionally, the IC design, OOC performance, the steady-state
ARLs for the HWMA_TBE control chart are also offered in this section.
The assessment of the control charts is often evaluated via run-length indicators, such as ARL, standard deviation run
length (SDRL), and percentile measures. The ARL is defined as the average number of sample points plotted until a
control chart detects an OOC signal.32 The ARLs of two types, that is, ARL0 and out of control ARL (ARL1 ). If the process
is operating in an IC state, the ARL0 is expected to be large in order to avoid frequent false alarms, while for OOC, ARL1
should be smaller to detect the process shift quickly.33 For improved performance of the control chart, it is essential to
have a smaller ARL1 for the control chart with fixed ARL0 at the desired level.12
In order to evaluate the overall assessment of the control chart, different performance indicators, such as extra quadratic
loss (EQL), relative average run length (RARL), and performance comparison index (PCI), are used. Mathematically, these
performance indicators are defined as follows

𝛿𝑚𝑎𝑥 ⎤
−1
EQL = (𝛿𝑚𝑎𝑥 − 𝛿𝑚𝑖𝑛 ) 𝛿 2 ARL (𝛿) 𝑑𝛿, ⎥
∫𝛿 ⎥
𝑚𝑖𝑛
𝛿𝑚𝑎𝑥 ⎥
−1 ARL (𝛿) ⎥
RARL = (𝛿𝑚𝑎𝑥 − 𝛿𝑚𝑖𝑛 ) 𝑑𝛿,⎥
∫𝛿 ARLBM (𝛿)
𝑚𝑖𝑛 ⎥
EQL ⎥
PCI = , ⎥
EQLBM ⎦

where 𝛿𝑚𝑖𝑛 and 𝛿𝑚𝑎𝑥 denote the minimum and maximum value of shift 𝛿. ARL(𝛿) is a competing control chart ARL
and ARLBM (𝛿) is a benchmark control chart ARL at shift 𝛿. A control chart with a smaller ARL at shift 𝛿 is called the
benchmark control chart. The EQL is a weighted average of ARL over a certain range of shift, that is, 𝛿𝑚𝑖𝑛 < 𝛿 < 𝛿𝑚𝑎𝑥
with 𝛿 2 is used as weight. It evaluates the control chart performance over the entire shift domain (𝛿𝑚𝑖𝑛 , 𝛿𝑚𝑎𝑥 ). A control
chart with a smaller EQL is regarded to be more superior. Similarly, the RARL is defined as the ratio between the ARL of
a particular control chart and the ARL of the benchmark control chart. Likewise, PCI is defined as the ratio of EQL of the
best control chart to the EQL of the benchmark control chart. The PCI also evaluates the performance of the best control
chart. The PCI = 1 for the benchmark control chart and the rest of the control charts have PCI > 1.
The current study is based on the smoothing parameter 𝜆, and the control limits width, that is, 𝐿. The combination (𝜆, 𝐿)
is chosen in such a way that the ARL0 is almost equal to 370. In addition, time-varying lower control limits are considered
for the control charts.

3.1 Run-length distribution of HWMA_TBE control chart

The run-length distribution of the control charts is generally calculated using three approaches, that is, the
integral equations, the Markov chain, and the Monte Carlo simulations. Vardeman and Ray4 and Gan5 implemented the
integral equations approach, whereas Borror, et al.,34 Liu, et al.,35 Zhang, et al.,36 and Hu, et al.37 used the Markov chain
approach to compute the ARL in their studies with asymptotic control limits. This study used the Monte Carlo simulations
8 ASLAM et al.

approach as a computational methodology to approximate the run-length distribution of the HWMA_TBE control chart.
The design structure of the HWMA_TBE control chart is complicated, and the time-varying control limits make it very
difficult to apply the first two approaches. The simulation algorithm for the numerical results can be explained in term of
several steps, given as follows

(i) Choose the combination of design parameters (𝜆, 𝐿) for a given 𝑘, and set the parameter 𝜃, that is, if the process is
IC, then 𝜃 = 𝜃0 = 1; and if the process is OOC, then 𝜃 = 𝜃1 = 𝛿𝜃0 with 0 < 𝛿 < 1.
(ii) Generate 106 random numbers from a Gamma(𝑘, 𝜃) and determine the 𝐿𝐶𝐿𝑡 , based on 𝜆 and 𝐿, using Equation (23).
(iii) Compute the plotting statistic 𝑈𝑡 , 𝑡 = 1, 2, … , 106 using Equation (18) with the starting value 𝑈0 = 𝑘.
(iv) Plot 𝑈𝑡 against 𝐿𝐶𝐿𝑡 . If 𝑈𝑡 > 𝐿𝐶𝐿𝑡 then the process is IC, but if 𝑈𝑡 ≤ 𝐿𝐶𝐿𝑡 , then the process is OOC. In this case,
stop the simulations and record the run length (RL).
(v) Repeat steps (ii) to (iv) 106 times and compute the approximate ARL and SDRL, using the respective expression,
given as follows

∑𝑁
𝑡=1 RL𝑡
ARL = ,
𝑁

∑𝑁 2
𝑡=1 (RL𝑡 − ARL)
SDRL = .
𝑁−1

Here, 𝑁 is the size of simulation runs, and in this study, it is set as 𝑁 = 106 .

3.2 IC design of HWMA_TBE control chart

In order to perform the simulation study for each 𝑘 =1, 2, 3, 4, the values of 𝜆 are chosen as; 𝜆 = 0.03, 0.05, 0.1, and 0.2.
Consequently, each combination of (𝑘, 𝜆) is then used to determine the values of the control chart coefficient, 𝐿, so that
ARL0 is approximately equal to the desired value. Table 1 reports the 𝐿 values computed to achieve the ARL0 is roughly
equal to 200, 370, and 500. Furthermore, Table 1 also presents the other IC run length characteristics such as IC SDRL
(SDRL0 ) and percentiles points, such as 𝑃5 , 𝑃25 , 𝑃50 , 𝑃75 , and 𝑃95 . Table 1 summarizes the following main points.

(i) For a given 𝜆, the value of 𝐿 increases as 𝑘 increases in order to achieve ARL0 approximately 370. For instance, for
the HWMA_TBE control chart with 𝜆 = 0.05, when 𝑘 = 1 then 𝐿 = 0.775; when 𝑘 = 2, then 𝐿 = 0.995; when 𝑘 = 3,
then 𝐿 = 1.117; and when 𝑘 = 4 then 𝐿 = 1.197 to attain ARL0 close to 370 (see Table 1).
(ii) For a specified 𝑘, as 𝜆 increases, 𝐿 also increases to attain ARL0 approximately 370. For instance, for the HWMA_TBE
control chart with 𝑘 = 3, when 𝜆 = 0.03 then 𝐿 = 0.913; when 𝜆 = 0.05 then 𝐿 = 1.117; when 𝜆 = 0.1 then 𝐿 = 1.355;
and when 𝜆 = 0.2 then 𝐿 = 1.498 to obtain ARL0 close to 370 (see Table 1).
(iii) The SDRL0 values for the HWMA_TBE control chart are greater than the corresponding ARL0 values. However, the
SDRL0 values decrease as 𝜆 and/or 𝑘 increase when ARL0 approximately 200. For instance, for the HWMA_TBE
control chart with (𝑘, 𝜆) = (1, 0.03), (2, 0.05), (3, 0.1), and (4, 0.2), the SDRL0 values are 1083.36, 603.42, 391.36, and
303.29, respectively (see Table 1).

3.3 OOC performance of HWMA_TBE control chart

To address the OOC performance of the HWMA_TBE control chart, for each 𝑘 = 1, 2, 3, 4, different design parameter
combinations (𝜆, 𝐿) are used, as presented in Table 1. As stated earlier, these combinations are set to achieve ARL0 close
to 370. For the case of the IC process, 𝜃 = 𝜃0 = 1, while for OOC process 𝜃 = 𝛿𝜃0 and 0 < 𝛿 < 1 as the only concern is to
monitor the downward shifts in the process. Following Chakraborty, et al.,11 for the OOC process, the values of 𝛿 are set
as; 𝛿 = 0.975, 0.95, 0.925, 0.9, 0.8, 0.85, 0.7, 0.5, and 0.25. The OOC performance results for the HWMA_TBE control chart
are given in Tables 2–5, which contribute to the following conclusions.
ASLAM et al. 9

TA B L E 1 IC run length characteristics and values of 𝐿 to achieve ARL0 ≅ 200, 370, and 500
𝒌=𝟏
𝝀 = 𝟎.𝟎𝟑 𝝀 = 𝟎.𝟎𝟓 𝝀 = 𝟎.𝟏 𝝀 = 𝟎.𝟐
𝐿 0.538 0.661 0.713 0.669 0.775 0.816 0.823 0.891 0.914 0.921 0.954 0.965
ARL0 200.35 370.44 499.29 199.27 369.89 499.56 200.53 370.44 499.56 200.48 370.33 500.19
SDRL0 1083.36 1961.20 2693.72 885.42 1730.96 2471.22 788.62 1673.90 2489.48 787.90 1777.22 2692.83
𝑃5 1 1 1 1 1 1 1 1 1 1 2 2
𝑃25 1 1 1 1 3 3 3 4 4 4 5 5
𝑃50 4 6 7 6 10 12 12 17 19 15 17 18
𝑃75 25 52 71 47 87 110 73 107 123 75 93 103
𝑃95 773 1529.1 2018.1 898.05 1630.05 2091 933 1601.05 2012.05 918 1537 1885.1
𝒌=𝟐
𝝀 = 𝟎.𝟎𝟑 𝝀 = 𝟎.𝟎𝟓 𝝀 = 𝟎.𝟏 𝝀 = 𝟎.𝟐
𝐿 0.662 0.826 0.899 0.84 0.995 1.224 1.07 1.182 1.225 1.228 1.293 1.315
ARL0 199.88 369.63 500.55 199.65 370.39 498.65 200.99 369.91 501.06 199.47 370.16 499.25
SDRL0 772.59 1306.34 1728.43 603.42 1096.14 1388.08 496.57 976.50 1394.29 449.44 931.72 1341.03
𝑃5 1 1 1 1 1 3 1 2 3 2 3 3
𝑃25 1 3 3 3 4 11 6 10 11 9 12 13
𝑃50 6 12 18 13 28 60 31 50 61 38 52 58
𝑃75 57 135 200 110 218 324 158 260 326 172 269 323
𝑃95 1027 1933 2619.05 1037 1909.05 2493 981 1856 2510.05 964 1876 2566.1
𝒌=𝟑
𝝀 = 𝟎.𝟎𝟑 𝝀 = 𝟎.𝟎𝟓 𝝀 = 𝟎.𝟏 𝝀 = 𝟎.𝟐
𝐿 0.725 0.913 1 0.931 1.117 1.197 1.208 1.355 1.413 1.408 1.498 1.533
ARL0 200.42 371.04 500.62 200.39 371.02 500.72 200.31 370.59 500.54 199.71 371.54 499.71
SDRL0 640.22 1066.28 1385.50 495.53 873.46 1159.86 391.36 754.20 1047.54 344.59 689.40 955.28
𝑃5 1 1 1 1 1 1 1 3 3 3 4 4
𝑃25 1 3 4 4 6 9 9 16 21 15 20 23
𝑃50 8 19 31 21 49 73 50 85 106 62 94 112
𝑃75 84 205 307 154 310 424 206 361 468 226 396 513
𝑃95 1094.05 2006.05 2671 1030 1858 2503 926 1748 2403 878 1707 2370
𝒌=𝟒
𝝀 = 𝟎.𝟎𝟑 𝝀 = 𝟎.𝟎𝟓 𝝀 = 𝟎.𝟏 𝝀 = 𝟎.𝟐
𝐿 0.765 0.968 1.067 0.989 1.197 1.291 1.299 1.469 1.54 1.525 1.635 1.678
ARL0 200.93 369.39 499.49 200.71 371.67 499.72 200.90 370.09 500.14 200.4 370.59 500.49
SDRL0 581.23 950.06 1215.66 443.11 762.38 1007.04 345.99 647.38 888.43 303.29 586.28 806.56
𝑃5 1 1 1 1 1 1 1 3 4 3 5 5
𝑃25 3 4 5 4 8 13 12 24 31 20 30 35
𝑃50 9 25 42 27 69 101 63 113 146 78 130 164
𝑃75 100 257 388 184 373 506 234 421 559 251 458 616
𝑃95 1116 1980 2622 1012 1788.05 2389.05 878 1625 2243 811 1550 2126

(i) For a given 𝜆, the OOC performance of the HWMA_TBE control chart boosts as 𝑘(𝑘 > 1) increases. For instance,
when 𝜆 = 0.05 and 𝛿 = 0.975, then the HWMA_TBE control chart the ARL1 and the OOC SDRL (SDRL1 ) values are
115.16 and 247.44 for 𝑘 = 2; 113.82 and 206.49 for 𝑘 = 3; and 110.29 and 179.61 for 𝑘 = 4, respectively (see Table 3–5).
(ii) For a given 𝑘, the OOC performance of the HWMA_TBE control chart deteriorates as 𝜆 increases. For instance, for
the HWMA_TBE control chart with 𝑘 = 2 and 𝛿 = 0.9, the ARL1 and SDRL1 values are 18.29 and 27.42 if 𝜆 = 0.03;
28.18 and 43.15 if 𝜆 = 0.05; 33.19 and 43.81 if 𝜆 = 0.1; and 34.21 and 44.71 if 𝜆 = 0.2, respectively (see Table 3).
10 ASLAM et al.

TA B L E 2 Run-length characteristics for HWMA_TBE control chart when 𝑘 = 1 and ARL0 ≅ 370
𝜹
𝝀 𝑳 1 0.975 0.95 0.925 0.9 0.85 0.8 0.7 0.5 0.25
0.03 0.661 ARL 370.44 96.06 50.18 32.04 23.49 14.16 9.80 5.70 2.88 1.56
SDRL 1961.20 326.02 143.73 80.17 54.03 28.03 17.26 7.90 2.66 1.00
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 1 1 1 1 1 1 1 1 1 1
𝑃50 6 5 5 4 4 4 3 3 2 1
𝑃75 52 37 28 22 18 13 10 6 4 2
𝑃95 1529.1 513 262 165 117 66 42 21 8 3
0.05 0.775 ARL 369.89 106.97 57.75 38.31 27.88 17.4 12.04 7.06 3.45 1.86
SDRL 1730.96 324.43 143.98 84.25 55.98 30.44 18.59 8.98 2.97 1.14
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 3 3 3 3 3 2 2 1 1 1
𝑃50 10 9 8 7 6 5 5 4 3 1
𝑃75 87 59 43 33 27 19 14 9 4 3
𝑃95 1630.05 551 291 187 130 75 49 25 9 4
0.1 0.891 ARL 370.44 110.37 61.27 41.57 30.7 19.57 13.8 8.27 4.14 2.28
SDRL 1673.9 314.14 140.16 82.52 55.68 29.94 18.8 9.17 3.14 1.16
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 4 4 4 3 3 3 3 3 2 1
𝑃50 17 14 12 11 10 8 7 5 3 2
𝑃75 107 72 52 41 33 23 17 10 5 3
𝑃95 1601.05 542 291 187 131 78 50 27 10 4
0.2 0.954 ARL 370.33 105.69 58.09 39.37 29.31 18.97 13.61 8.33 4.31 2.43
SDRL 1777.22 312.94 136.41 79.04 52.48 28.45 17.67 8.62 3.00 0.99
𝑃5 2 2 2 2 2 1 1 1 1 1
𝑃25 5 4 4 4 4 4 3 3 2 2
𝑃50 17 15 13 11 10 8 7 5 3 2
𝑃75 93 65 48 38 31 22 17 11 5 3
𝑃95 1537 514 270 174 123 72 48 25 10 4

3.4 Steady-State 𝐀𝐑𝐋

The ARLs calculated in the previous section are called the zero-state ARLs, as the downward shift is assumed to be
occurred at the starting phase of the process, that is, at time 𝑡 = 1. However, it is important to evaluate the steady-state
ARLs, where the shift may not occur at the beginning phase of the process. In this case, for each 𝑘 and 𝜆 = 0.05 and 0.1,
the steady-state ARLs for the HWMA_TBE control chart are computed. The shift is assumed to be occurred at times 𝑡 =
50 , 100, 200, and 300. The steady-state ARLs along with the zero-state ARLs are demonstrated in Table 6, and the related
conclusions are given as follows

(i) The steady-state ARLs are larger than the respective zero-state ARLs, and as 𝑡 increases, the steady-state ARLs also
increase (see Table 6). This discrepancy is may be due to time-varying control limits, which are narrow at the starting
phase of the process, so the steady-state ARLs are higher than the respective zero-state ARLs.
(ii) For small shifts, that is,𝛿 ≥ 0.9, the steady-state ARLs are close to their respective zero-state ARLs. However, for
moderate and large shifts, the steady-state ARLs are higher than their respective zero-state ARLs (see Table 6). For
example, with 𝑘 = 1 and 𝜆 = 0.1, when 𝛿 = 0.975 then the steady-state ARLs are 109.95, 110.03, 110.39, 110.48 for
𝑡 = 50, 100, 200, 300, respectively, which are 0.01%, 0.08%, 0.41%, 0.49% larger than the respective zero-state ARL
of 109.94, whereas when 𝛿 = 0.8 then these values are 13.96, 14.04, 14.46, 14.71 for 𝑡 = 50, 100, 200, 300, respectively,
ASLAM et al. 11

TA B L E 3 Run-length characteristics for HWMA_TBE control chart when 𝑘 = 2 and ARL0 ≅ 370
𝜹
𝝀 𝑳 1 0.975 0.95 0.925 0.9 0.85 0.8 0.7 0.5 0.25
0.03 0.826 ARL 369.39 90.15 44.01 26.98 18.29 10.51 6.98 4.01 2.00 1.06
SDRL 950.06 179.07 77.66 43.74 27.42 13.7 7.91 3.59 1.34 0.34
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 4 3 3 3 3 3 1 1 1 1
𝑃50 25 15 11 8 7 5 4 3 1 1
𝑃75 257 91 50 32 22 13 9 5 3 1
𝑃95 1980 441 199 116 75 39 23 11 4 1
0.05 0.995 ARL 370.39 115.16 61.51 39.63 28.18 16.84 11.39 6.46 3.13 1.59
SDRL 1096.14 247.44 112.12 65.47 43.15 22.84 13.81 6.42 2.13 0.92
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 4 4 4 4 3 3 3 3 1 1
𝑃50 28 21 16 13 10 8 6 4 3 1
𝑃75 218 108 68 47 34 22 15 8 4 3
𝑃95 1909.05 557 279 169 115 63 40 19 7 3
0.1 1.182 ARL 369.91 124.45 69.21 45.51 33.19 20.27 13.99 8.11 3.90 2.10
SDRL 976.5 240.13 111.88 65.51 43.81 23.47 14.48 6.93 2.25 0.98
𝑃5 2 2 2 1 1 1 1 1 1 1
𝑃25 10 8 7 6 6 5 4 3 3 1
𝑃50 50 34 26 20 17 12 9 6 3 2
𝑃75 260 129 82 57 43 27 19 11 5 3
𝑃95 1856 557 285 174 121 67 43 22 8 3
0.2 1.293 ARL 370.16 129.02 71.13 46.68 34.21 20.87 14.65 8.53 4.15 2.32
SDRL 931.72 248.63 115.07 66.86 44.71 23.40 14.57 6.83 2.19 0.79
𝑃5 3 3 3 3 3 2 2 2 2 1
𝑃25 12 10 9 8 7 6 5 4 3 2
𝑃50 52 36 27 21 18 13 10 6 4 2
𝑃75 269 130 82 57 43 27 19 11 5 3
𝑃95 1876 582 294 178 123 67 43 22 8 3

which are 0.65%, 1.23%, 4.25%, 6.06% larger than the respective zero state ARL of 13.87. This happened due to the
same reason as above.
(iii) For a given 𝜆, the steady-state performance of the HWMA_TBE control chart enhances as 𝑘(𝑘 > 1) increases. For
instance, with 𝑡 = 50, 𝜆 = 0.05, and 𝛿 = 0.925, the steady-state ARLs for the HWMA_TBE control chart are 39.82,
38.08, 35.59 for 𝑘 = 2, 3, and 4, respectively (see Table 6).
(iv) For a specified 𝑘, the steady-state performance of the HWMA_TBE control chart deteriorates as 𝜆 increases. For
example, with 𝑡 = 50, 100, 200, 300, 𝑘 = 2 and 𝛿 = 0.9, the steady-state ARLs for the HWMA_TBE control chart are
28.39, 28.59, 28.95, 29.44 for 𝜆 = 0.05, and for 𝜆 = 0.1 these values are 33.33, 33.49, 33.84, 34.25 (see Table 6).

4 PHASE-II HWMA_TBE control chart

This section addresses the run length distribution of the Phase-II HWMA_TBE control chart, where the parameter 𝜃 is
assumed to be unknown (referred to as Case U). Furthermore, this section also presents the design structure of the Phase-
II HWMA TBE control chart. Further, this investigates the impact of the parameter estimation on the performance of the
HWMA_TBE control chart. In Phase-II monitoring, the control chart is implemented to monitor the potential shifts from
the IC state. In practice, the IC value of the parameter 𝜃, that is, 𝜃0 , is generally not known in advance, particularly at
the beginning phase of the process. In this case, one choice is to estimate 𝜃 using an appropriate estimator based on the
12 ASLAM et al.

TA B L E 4 Run-length characteristics for HWMA_TBE control chart when 𝑘 = 3 and ARL0 ≅ 370
𝜹
𝝀 𝑳 1 0.975 0.95 0.925 0.9 0.85 0.8 0.7 0.5 0.25
0.03 0.931 ARL 371.04 95.48 47.29 29.02 20.39 11.63 7.77 4.44 2.23 1.14
SDRL 1066.28 208.53 89.61 50.47 32.56 16.19 9.57 4.28 1.55 0.51
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 3 3 3 3 3 3 1 1 1 1
𝑃50 19 13 10 8 7 5 4 3 1 1
𝑃75 205 85 49 32 24 14 10 6 3 1
𝑃95 2006.05 483 225 130 86 45 27 13 5 3
0.05 1.117 ARL 371.02 113.82 59.70 38.00 26.59 15.61 10.35 5.80 2.80 1.36
SDRL 873.46 206.49 94.27 55.06 36.01 18.77 11.22 5.06 1.74 0.74
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 6 5 5 4 4 3 3 3 1 1
𝑃50 49 29 21 16 12 8 6 4 3 1
𝑃75 310 129 74 49 35 21 14 8 4 1
𝑃95 1858 513 250 149 100 54 33 16 6 3
0.1 1.355 ARL 370.59 128.03 69.97 45.33 32.71 19.48 13.31 7.44 3.55 1.84
SDRL 754.20 203.65 96.03 56.30 37.37 19.60 12.11 5.62 1.79 0.90
𝑃5 3 3 3 3 2 2 2 1 1 1
𝑃25 16 13 10 8 7 6 5 3 3 1
𝑃50 85 49 34 25 20 13 10 6 3 2
𝑃75 361 155 90 60 44 27 18 10 4 3
𝑃95 1748 526 262 159 108 59 38 19 7 3
0.2 1.498 ARL 371.54 138.55 75.56 49.25 35.14 20.94 14.29 8.11 3.85 2.11
SDRL 689.40 215.70 103.23 61.05 39.38 20.49 12.48 5.76 1.78 0.75
𝑃5 4 3 3 3 3 3 3 2 2 1
𝑃25 20 16 12 10 9 7 5 4 3 2
𝑃50 94 54 37 27 22 14 10 7 3 2
𝑃75 396 168 96 64 47 28 19 11 5 3
𝑃95 1707 564 282 172 114 62 39 19 7 3

IC Phase-I sample of size 𝑚. In this study, the maximum likelihood (ML) estimator is used to estimate the parameter 𝜃,
which is the uniformly minimum variance unbiased estimator.11,17
iid
Let 𝑋𝑡 ∼ Gamma(𝑘, 𝜃), 𝑡 = 1, 2, … , 𝑚, denote the IC Phase-I sample of size 𝑚, then based on this sample, the ML esti-
mate of 𝜃 is given as follows
∑𝑚
𝑋̄ 𝑡=1 𝑋𝑡
𝜃̂ = = . (24)
𝑘 𝑘𝑚

This ML estimate of 𝜃, that is, 𝜃̂ is then used to compute the statistic 𝑈𝑡 and estimate the control limits 𝑈𝐶𝐿𝑡 and 𝐿𝐶𝐿𝑡
of the Phase-II HWMA_TBE control chart. The plotting statistic 𝑈𝑡 , in this case, is defined as follows
]
̂ 𝑡 = 1,
𝑈𝑡 = 𝜆𝑋𝑡 + (1 − 𝜆) 𝑘𝜃,
(25)
𝑈𝑡 = 𝜆𝑋𝑡 + (1 − 𝜆) 𝑋̄ 𝑡−1 , 𝑡 > 1.

The estimated IC mean and variance of 𝑈𝑡 based on the IC Phase-I sample are, respectively, defined as

ˆ
𝐸 (𝑈𝑡 |IC Phase−I) = 𝑘𝜃, (26)
ASLAM et al. 13

TA B L E 5 Run-length characteristics for HWMA_TBE control chart when 𝑘 = 4 and ARL0 ≅ 370
𝜹
𝝀 𝑳 1 0.975 0.95 0.925 0.9 0.85 0.8 0.7 0.5 0.25
0.03 0.968 ARL 369.39 90.15 44.01 26.98 18.29 10.51 6.98 4.01 2.00 1.06
SDRL 950.06 179.07 77.66 43.74 27.42 13.70 7.91 3.59 1.34 0.34
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 4 3 3 3 3 3 1 1 1 1
𝑃50 25 15 11 8 7 5 4 3 1 1
𝑃75 257 91 50 32 22 13 9 5 3 1
𝑃95 1980 441 199 116 75 39 23 11 4 1
0.05 1.197 ARL 371.67 110.29 56.99 35.71 24.59 14.20 9.46 5.23 2.53 1.20
SDRL 762.38 179.61 82.43 47.89 31.01 15.87 9.49 4.22 1.51 0.57
𝑃5 1 1 1 1 1 1 1 1 1 1
𝑃25 8 6 5 5 4 4 3 3 1 1
𝑃50 69 36 23 17 13 8 6 4 3 1
𝑃75 373 136 75 48 33 19 13 7 3 1
𝑃95 1788.05 469 225 133 88 47 29 14 5 3
0.1 1.469 ARL 370.09 126.21 68.19 43.77 31.05 18.30 12.32 6.81 3.26 1.59
SDRL 647.38 178.63 84.36 49.81 32.66 17.05 10.51 4.71 1.56 0.81
𝑃5 3 3 3 3 3 2 2 1 1 1
𝑃25 24 16 12 10 8 6 5 3 2 1
𝑃50 113 58 38 27 20 13 9 6 3 1
𝑃75 421 162 92 59 43 25 17 9 4 2
𝑃95 1625 481 237 144 96 53 33 16 6 3
0.2 1.635 ARL 370.59 140.28 76.17 48.70 34.27 20.22 13.56 7.53 3.54 1.87
SDRL 586.28 193.95 94.24 54.95 35.57 18.25 10.98 4.94 1.51 0.73
𝑃5 5 4 4 3 3 3 3 2 1 1
𝑃25 30 20 15 12 10 7 6 4 3 1
𝑃50 130 66 42 30 23 15 10 6 3 2
𝑃75 458 181 101 65 46 27 18 10 4 2
𝑃95 1550 531 265 159 105 56 35 17 6 3

⎧ 𝜆2 𝑘𝜃ˆ2 ] if 𝑡 = 1,
⎪[
Var (𝑈𝑡 |IC Phase−I) = ⎨ 2 (1−𝜆)2 ˆ2 (27)
⎪ 𝜆 + 𝑡−1 𝑘𝜃 if 𝑡 > 1.

In the same lines, for the two-sided Phase-II HWMA_TBE control chart, the estimated time-varying control limits
ˆ
denoted by 𝑈 ˆ 𝑡 , respectively, and given as follows
𝐶𝐿𝑡 and 𝐿𝐶𝐿

⎧ √ ⎤
⎪𝑘𝜃̂ + 𝐿 𝜆2 𝑘𝜃̂ 2 if 𝑡 = 1, ⎥
⎪ √[ ⎥
ˆ
𝑈 𝐶𝐿𝑡 = ⎨ 2
] ⎥
(1−𝜆)
⎪𝑘𝜃̂ + 𝐿 𝜆2 + 𝑘𝜃̂ 2 if 𝑡 > 1, ⎥
⎪ 𝑡−1 ⎥
⎩ ⎥
⎥ (28)
⎧ √ ⎥
⎪𝑘𝜃̂ − 𝐿 𝜆2 𝑘𝜃̂ 2 if 𝑡 = 1, ⎥
⎪ √[ ⎥
ˆ𝑡 = ⎨
𝐿𝐶𝐿 2
] ⎥
(1−𝜆)
⎪𝑘𝜃̂ − 𝐿 𝜆2 + 𝑘𝜃̂ 2 if 𝑡 > 1. ⎥
⎪ 𝑡−1 ⎥
⎩ ⎦
14 ASLAM et al.

TA B L E 6 Steady-state ARLs versus zero-state ARLs for HWMA_TBE control chart


𝜹
𝒌 𝝀 𝑳 𝒕 0.975 0.95 0.925 0.9 0.85 0.8 0.7 0.5 0.25
1 0.05 0.775 1 106.77 58.35 38.60 28.09 17.29 12.02 7.04 3.47 1.86
50 106.79 58.44 38.68 28.19 17.40 12.11 7.13 3.56 1.95
100 106.82 58.51 38.76 28.27 17.47 12.19 7.22 3.64 2.03
200 107.08 58.90 39.11 28.66 17.88 12.61 7.63 4.06 2.47
300 107.16 59.13 39.34 28.91 18.13 12.89 7.90 4.34 2.74
0.1 0.890 1 109.94 61.85 41.52 30.80 19.45 13.87 8.29 4.13 2.27
50 109.95 61.92 41.61 30.88 19.54 13.96 8.38 4.22 2.36
100 110.03 61.98 41.67 30.96 19.62 14.04 8.46 4.30 2.45
200 110.39 62.35 42.08 31.36 20.01 14.46 8.87 4.73 2.88
300 110.48 62.50 42.36 31.58 20.25 14.71 9.14 4.99 3.14
2 0.05 0.995 1 114.52 61.58 39.70 28.26 16.85 11.41 6.45 3.12 1.60
50 114.69 61.72 39.82 28.39 16.99 11.55 6.59 3.27 1.74
100 114.78 61.88 39.98 28.59 17.20 11.77 6.82 3.49 1.97
200 115.02 62.29 40.37 28.95 17.57 12.12 7.19 3.87 2.35
300 115.42 62.67 40.82 29.44 18.06 12.65 7.71 4.40 2.87
0.1 1.182 1 123.93 68.95 45.43 33.17 20.27 14.03 8.05 3.89 2.10
50 124.00 69.07 45.60 33.33 20.43 14.20 8.21 4.05 2.27
100 124.09 69.21 45.74 33.49 20.62 14.39 8.41 4.26 2.47
200 124.35 69.55 46.09 33.84 20.97 14.76 8.78 4.63 2.85
300 124.73 69.92 46.53 34.25 21.40 15.21 9.25 5.11 3.32
3 0.05 1.117 1 113.86 59.81 37.96 26.48 15.50 10.38 5.78 2.79 1.36
50 113.98 59.95 38.08 26.63 15.65 10.52 5.92 2.93 1.50
100 114.03 60.11 38.27 26.83 15.85 10.72 6.13 3.14 1.71
200 114.34 60.45 38.66 27.23 16.26 11.14 6.55 3.57 2.13
300 114.75 60.87 39.04 27.65 16.69 11.58 7.00 4.03 2.59
0.1 1.355 1 127.69 69.85 45.43 32.56 19.53 13.28 7.47 3.57 1.84
50 127.80 69.96 45.56 32.70 19.67 13.42 7.60 3.71 1.99
100 127.85 70.08 45.72 32.85 19.83 13.59 7.78 3.88 2.17
200 128.16 70.43 46.05 33.20 20.20 13.97 8.16 4.27 2.55
300 128.38 70.72 46.39 33.52 20.54 14.34 8.55 4.66 2.94
4 0.05 1.197 1 111.31 56.94 35.45 24.57 14.18 9.42 5.22 2.54 1.20
50 111.39 57.10 35.59 24.72 14.35 9.60 5.39 2.71 1.37
100 111.46 57.24 35.75 24.89 14.52 9.78 5.57 2.89 1.56
200 111.85 57.69 36.21 25.31 14.94 10.22 6.03 3.35 2.02
300 112.00 57.92 36.50 25.57 15.23 10.51 6.32 3.64 2.32
0.1 1.469 1 127.37 68.26 43.68 30.83 18.26 12.32 6.83 3.26 1.60
50 127.48 68.41 43.82 30.97 18.39 12.46 6.98 3.41 1.75
100 127.57 68.59 43.98 31.14 18.57 12.63 7.16 3.59 1.93
200 127.98 69.00 44.39 31.55 18.99 13.06 7.60 4.03 2.38
300 128.08 69.23 44.63 31.81 19.26 13.34 7.88 4.33 2.67

The factor 𝐿 denotes the control chart width, and whenever 𝐿𝐶𝐿ˆ 𝑡 computed is negative, then it is reset as; 𝐿𝐶𝐿
ˆ 𝑡 = 0. If
the only concern is to monitor the decreasing shifts in the process so the lower control limit of the HWMA_TBE control
ASLAM et al. 15

FIGURE 1 ARL values for the Phase-II HWMA_TBE control chart with various sets of design parameters

chart can only be used for this purpose, which is given as follows

⎧ √
⎪𝑘𝜃̂ − 𝐿 𝜆2 𝑘𝜃̂ 2 if 𝑡 = 1,
⎪ √[
ˆ𝑡 = ⎨
𝐿𝐶𝐿 2
] (29)
(1−𝜆)
⎪𝑘𝜃̂ − 𝐿 𝜆2 + 𝑘𝜃̂ 2 if 𝑡 > 1.
⎪ 𝑡−1

ˆ 𝑡 , and the process is considered to IC if 𝑈𝑡 > 𝐿𝐶𝐿


The statistic 𝑈𝑡 is plotted against the 𝐿𝐶𝐿 ˆ 𝑡 and if 𝑈𝑡 ≤ 𝐿𝐶𝐿
ˆ 𝑡 then the
process is known to be OOC.
The estimate 𝜃̂ is substituted in Equations (25), (26), and (27) in place of unknown parameter 𝜃 for the Case K, so, the
ˆ 𝑡 are random variables, and the Phase-II HWMA_TBE control chart’s performance depends on
initial value 𝑈0 and 𝐿𝐶𝐿
the Phase-I sample size 𝑚. In addition, it is important to investigate whether the design parameters 𝜆 and 𝐿 for Case K of
the HWMA_TBE control chart can be used to Case U, that is, the effect of parameter estimation on Phase-II HWMA_TBE
control chart. In this regard, the various combination of design parameters are considered as follows;

(i) (𝑘 = 1, 𝜆 = 0.05, 𝐿 = 0.775), (𝑘 = 1, 𝜆 = 0.1, 𝐿 = 0.890),


(ii) (𝑘 = 2, 𝜆 = 0.05, 𝐿 = 0.995), (𝑘 = 2, 𝜆 = 0.1, 𝐿 = 1.182).

For each design parameter combination, the IC Phase-I sample of sizes 𝑚 = 50, 100, 500, and 1000 are chosen. The
numerical results for the Phase-II HWMA_TBE control chart are generated using the same simulation algorithm as in
the preceding section. Figure 1 shows the results of the Phase-II HWMA_TBE control chart. These results may lead to the
following conclusions.
16 ASLAM et al.

TA B L E 7 The values of 𝐿 for the Phase-II HWMA_TBE control chart in Case U


𝒎
𝒌 𝝀 50 100 500 1000 Case K
1 0.05 0.652 0.709 0.755 0.767 0.775
0.1 0.821 0.853 0.878 0.885 0.890
2 0.05 0.834 0.908 0.968 0.984 0.995
0.1 1.085 1.130 1.165 1.175 1.182

(i) For each 𝑘, Case U ARL values are higher than respective Case K ARL values for small shifts. However, for moderate-
to-large shifts, that is, 𝛿 ≤ 0.9, the differences among the Case U and Case K ARL values are become negligible (see
Figure 1).
(ii) The differences among Case U and Case K ARL values reduce as the IC Phase-I sample size 𝑚 increases. In fact, for
larger shifts, that is, 𝛿 ≤ 0.5, Case K ARL values are equal to Case U ARL values are for all 𝑚 values (see Figure 1).

These findings contribute to the fact that design parameters, that is, (𝜆, 𝐿) of Case K for the HWMA_TBE control chart
cannot be implemented to Case U. In order to establish a more realistic method for the construction of the Phase-II
HWMA_TBE control chart, the values of 𝐿 are computed, using same simulation algorithm in the preceding section,
so that Case U ARL0 values are close to 370. Table 7 presents the results in this regard, which show that as the IC sample
size 𝑚 increases, the 𝐿 values for Case U converge to the relevant 𝐿 values for Case K.

5 COMPARATIVE STUDY

This section covers the comparative study of the HWMA_TBE control chart to some competing control charts, such as the
EWMA_TBE, DEWMA_TBE, TEWMA_TBE, DGWMA_TBE, and GWMA_TBE control charts. At a particular shift, the
performance indicators ARL and SDRL are used to compare are the HWMA_TBE control chart against the competing con-
trol chart, while the performance indicators EQL, RARL, and PCI evaluate the overall performance of the HWMA TBE
control chart. Moreover, for each 𝑘 = 1, 2, 3, the values of 𝜆 are taken as; 0.05 and 0.1 for the competing EWMA_TBE,
DEWMA_TBE, TEWMA_TBE control charts and the two combinations of design parameters (𝑞 = 0.8, 𝑎 = 0.7) and
(𝑞 = 0.9, 𝑎 = 0.7) are chosen for DGWMA_TBE and GWMA_TBE control charts for comparison with ARL0 = 370. The
numerical results for ARL and SDRL indicators for the competing control chart are shown in Tables 8–10, while the EQL,
RARL, and PCI values are presented in Table (11). More details about the comparative study are provided as follows

(i) HWMA_TBE Versus EWMA_TBE: The HWMA_TBE control chart provides excellent performance against the
EWMA_TBE control chart in terms of smaller ARL1 and SDRL1 values. For example, the HWMA_TBE control chart
with 𝑘 = 1, 𝜆 = 0.05 and 𝛿 = 0.95 delivers the ARL1 and SDRL1 values as 57.75 and 143.98, whereas the EWMA_TBE
control chart provide the ARL1 and SDRL1 values as 214.82 and 212.13, respectively (see Table 2 vs 8). Similarly,
the EQL, RARL, and PCI indicators show that the HWMA_TBE control chart outperforms the EWMA_TBE control
chart in terms of overall performance. As for 𝑘 = 1 and 𝜆 = 0.05, the EQL, RARL, and PCI values of the HWMA_TBE
control chart (ie, 16.0922, 1.0000, and 1.0000) are less than the EQL, RARL, and PCI values of the EWMA_TBE control
chart (ie, 43.8252, 2.7234, and 4.1664) (see Table 11).
(ii) HWMA_TBE Versus DEWMA_TBE: The HWMA_TBE control chart achieves outstanding performance against
the DEWMA_TBE control chart in terms of minimal ARL1 and SDRL1 values. For instance, the ARL1 and SDRL1
values for the HWMA_TBE control chart with 𝑘 = 2 and 𝜆 = 0.1 and 𝛿 = 0.95 are reported as 69.21 and 111.88, while
the ARL1 and SDRL1 values for the DEWMA_TBE control chart are observed as 178.31 and 175.31, respectively (see
Table 3 vs 9). Correspondingly, the EQL, RARL, and PCI values of the HWMA_TBE control chart reveal an edge
in the overall performance against the DEWMA_TBE control chart. For example, with 𝑘 = 2 and 𝜆 = 0.1, the EQL,
RARL, and PCI values for the HWMA_TBE are 26.1034, 1.0000, and 1.0000, while these values for the DEWMA_TBE
control charts are 57.3706, 2.1978, and 2.1684 (see Table 11).
(iii) HWMA_TBE Versus TEWMA_TBE: The HWMA_TBE control chart outperforms the TEWMA_TBE control chart
in terms of lower ARL1 and SDRL1 values. For instance, with 𝑘 = 3, 𝜆 = 0.05 and 𝛿 = 0.925, the HWMA_TBE
control chart delivers the ARL1 and SDRL1 values of 38.00 and 55.06, whereas the TEWMA_TBE control chart
ASLAM et al. 17

TA B L E 8 Run-length characteristics for competing control charts when 𝑘 = 1 and ARL0 ≅ 370
𝐓𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐆𝐖𝐌𝐀_𝐓𝐁𝐄 𝐆𝐖𝐌𝐀_𝐓𝐁𝐄
𝒒 = 𝟎.𝟖, 𝒒 = 𝟎.𝟗, 𝒒 = 𝟎.𝟖, 𝒒 = 𝟎.𝟗,
𝝀 = 0.05 0.1 0.05 0.1 0.05 0.1 𝒂 = 0.7 𝒂 = 0.7 𝒂 = 0.7 𝒂 = 0.7
𝜹 𝑳 =1.247 1.550 1.460 1.724 1.868 1.911 1.846 1.510 1.750 1.806
1 370.37 370.35 370.25 370.10 370.79 370.00 370.08 369.74 370.06 370.96
(411.27) (375.82) (396.12) (370.80) (373.53) (362.35) (373.05) (418.73) (353.90) (356.76)
0.975 259.55 278.92 265.09 284.61 281.01 295.48 278.53 253.63 288.11 270.12
(284.03) (279.12) (280.34) (283.18) (283.51) (289.90) (277.70) (286.44) (281.96) (259.63)
0.95 188.87 212.87 195.55 219.60 214.82 237.13 209.42 178.6 228.45 202.60
(204.21) (210.93) (204.41) (216.18) (212.13) (232.61) (204.84) (198.66) (217.78) (190.19)
0.925 140.90 164.62 147.69 172.46 166.21 191.02 163.46 130.97 183.16 157.98
(147.99) (159.96) (150.61) (167.50) (160.52) (186.50) (158.86) (142.32) (172.63) (140.88)
0.9 108.41 129.59 113.46 137.35 131.44 154.84 129.06 99.23 148.58 125.37
(110.08) (123.08) (112.83) (131.03) (124.23) (147.90) (122.82) (104.70) (135.05) (108.34)
0.85 68.79 83.88 71.93 89.36 85.81 105.15 85.61 64.12 100.97 83.87
(65.58) (74.87) (67.32) (80.88) (76.33) (96.25) (76.01) (64.79) (88.18) (65.30)
0.8 47.43 57.79 49.18 61.22 59.15 73.16 58.26 42.49 70.97 60.14
(42.63) (47.97) (42.99) (52.01) (48.21) (63.77) (47.91) (39.03) (58.20) (43.04)
0.7 26.13 31.82 26.68 32.86 32.44 39.22 31.97 22.97 39.70 35.55
(21.40) (22.27) (20.43) (23.81) (22.68) (30.15) (22.62) (18.18) (27.46) (43.04)
0.5 10.66 13.81 11.05 13.68 13.95 15.55 13.68 9.75 17.05 17.29
(7.36) (7.10) (6.50) (6.78) (6.67) (7.95) (6.59) (5.69) (7.63) (6.22)
0.25 4.55 6.53 5.03 6.53 6.76 7.23 6.60 4.65 8.53 9.55
(1.87) (1.97) (1.64) (1.71) (1.65) (1.77) (1.63) (1.40) (1.89) (1.69)

produce the ARL1 and SDRL1 values of 81.72 and 86.63, respectively (see Table 4 vs 10). In addition, over a cer-
tain range of shifts, the EQL, RARL, and PCI values demonstrate the HWMA_TBE control chart’s superiority
over the TEWMA_TBE control chart. For instance, with 𝑘 = 3 and 𝜆 = 0.05 the HWMA_TBE control chart pro-
vides EQL = 15.9537, RARL = 1.0000, and PCI = 1.0000, while the TEWMA_TBE control chart offers EQL =
25.2745, RARL = 1.5842, and PCI = 1.6457, (see Table 11). This indicates the superior performance for the
HWMA_TBE control chart over the TEWMA_TBE control chart.
(iv) HWMA_TBE Versus DGWMA_TBE: The HWMA_TBE control chart reveals superior performance over the
DGWMA_TBE control chart in terms of least ARL1 and SDRL1 values. For instance, with 𝑘 = 2, 𝜆 = 0.05, 𝑞 = 0.8,
𝑎 = 0.7 and 𝛿 = 0.95, the ARL1 and SDRL1 values for the HWMA_TBE control chart are observed as; 61.51 and
112.12, where the ARL1 and SDRL1 values for the DGWMA_TBE control chart result in 172.63 and 168.60, respec-
tively (see Table 3 vs 9). Moreover, the EQL, RARL, and PCI values indicate improved overall performance for the
HWMA_TBE control chart as compared to the DGWMA_TBE control chart. For example, for control chart param-
eters stated above, the DGWMA_TBE control chart has larger EQL, RARL, and PCI (ie, 33.7990, 2.0591, and 2.7138)
than the EQL, RARL, and PCI (ie, 16.4142, 1.0000, and 1.0000) of the HWMA_TBE control chart (see Table 11).
(v) HWMA_TBE Versus GWMA_TBE: The HWMA_TBE control chart delivers better performance against the
GWMA_TBE control chart in terms of small ARL1 and SDRL1 values. For instance, the ARL1 and SDRL1 values
for the HWMA_TBE control chart with 𝑘 = 3, 𝜆 = 0.1, 𝑞 = 0.9, 𝑎 = 0.7 and 𝛿 = 0.925, are 45.33 and 56.30, while the
ARL1 and SDRL1 values for the GWMA_TBE control chart are 103.33 and 88.10, respectively (see Table 4 vs 10). Also,
the overall performance indicators suggest that the HWMA_TBE control chart provides superior overall performance
against the GWMA_TBE control chart. For instance, with the same control charts properties, the EQL, RARL, and
PCI values of the HWMA_TBE control chart; 26.3317, 1.0000, and 1.0000, are smaller than the EQL, RARL, and PCI
values of the GWMA_TBE control chart; 56.0872, 2.1300, and 2.1369, respectively (see Table 11).
18 ASLAM et al.

TA B L E 9 Run-length characteristics for competing control charts when 𝑘 = 2 and ARL0 ≅ 370
𝐓𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐆𝐖𝐌𝐀_𝐓𝐁𝐄 𝐆𝐖𝐌𝐀_𝐓𝐁𝐄
𝒒 = 𝟎.𝟖, 𝒒 = 𝟎.𝟗, 𝒒 = 𝟎.𝟖, 𝒒 = 𝟎.𝟗,
𝝀 =0.05 0.1 0.05 0.1 0.05 0.1 𝒂 = 0.7 𝒂 = 0.7 𝒂 = 0.7 𝒂 = 0.7
𝜹 𝑳 =1.274 1.917 1.487 1.775 1.953 2.050 1.920 1.535 1.953 1.960
1 370.52 370.35 370.29 370.44 369.79 370.27 370.04 370.55 370.05 371.34
(427.47) (381.96) (408.17) (376.05) (377.59) (366.92) (380.94) (438.39) (360.38) (365.67)
0.975 224.07 246.83 230.43 253.79 250.66 269.50 247.47 214.53 267.09 245.81
(252.12) (250.16) (245.80) (254.20) (251.51) (263.51) (250.78) (247.55) (257.57) (232.48)
0.95 146.63 170.57 153.15 178.31 174.57 199.15 172.63 136.66 196.56 170.71
(161.34) (169.09) (161.63) (175.31) (169.48) (192.80) (168.60) (151.15) (187.96) (155.19)
0.925 101.03 122.33 106.78 129.55 124.96 149.22 122.93 93.06 146.92 124.04
(107.08) (118.38) (109.24) (125.61) (117.63) (141.95) (115.28) (100.10) (135.45) (108.21)
0.9 73.51 89.96 77.38 96.38 93.15 114.19 91.82 66.35 112.63 93.96
(75.10) (84.00) (76.76) (91.29) (85.52) (107.40) (84.29) (69.66) (100.70) (76.64)
0.85 43.79 53.69 45.43 56.63 55.27 69.28 54.34 38.82 69.59 59.01
(42.23) (46.32) (42.00) (49.72) (46.87) (61.38) (45.96) 37.98 (57.74) (42.68)
0.8 29.09 35.58 29.76 37.02 36.29 44.46 35.74 25.27 46.60 40.53
(26.80) (28.15) (25.78) (30.02) (28.43) (36.41) (28.09) (23.00) (35.07) (25.78)
0.7 15.18 18.97 15.46 19.11 19.05 22.39 18.73 13.23 24.95 23.40
(13.23) (14.86) (11.98) (12.92) (12.47) (15.43) (12.38) (10.49) (15.28) (11.81)
0.5 5.69 7.84 6.05 7.82 8.05 9.01 7.89 5.43 10.79 11.39
(4.24) (4.30) (3.65) (3.81) (3.65) (4.04) (3.64) (3.07) (4.24) (3.70)
0.25 2.28 3.47 2.73 3.68 3.93 4.30 3.83 2.64 5.50 6.28
(1.04) (1.16) (0.89) (0.97) (0.91) (0.99) (0.90) (0.76) (1.08) (1.00)

6 ROBUSTNESS STUDY

This section investigates the robustness study of the HWMA_TBE control chart. In this study, it is assumed that the mean
values of TBE data are known, and the goal is to monitor the time until only one failure occurs, that is, 𝑘 = 1; however, the
assumption of the gamma distribution for the TBE observations is violated. The robustness study considered the Weibull
and log-normal distributions, whereas the HWMA_TBE control chart is constructed for TBE data follows the gamma
distribution. More details about the robustness study are provided in the following subsections.

6.1 Weibull distribution

The Weibull distribution is a lifetime distribution having attractive properties and potential applications in many fields.
The two-parameter Weibull distribution for a random variable 𝑋 can be defined by the pdf, given as follows
( ( )𝛽 )
𝛽 ( 𝑥 )𝛽−1 𝑥
𝑓 (𝑥; 𝛼, 𝛽) = exp − , 𝑥, 𝛼, 𝛽 > 0, (30)
𝛼 𝛼 𝛼

where 𝛼 and 𝛽 denote the scale and shape parameters, respectively. The Weibull distribution can often be represented
by the notation as Weibull(𝛼, 𝛽). The Weibull distribution has a failure rate that decreases for 𝛽 < 1, increases for 𝛽 > 1,
and remains constant for 𝛽 = 1. The Weibull distribution can be referred to as an exponential distribution when its shape
parameter 𝛽 = 1; however, it is reduced to the Rayleigh distribution when 𝛽 = 2.
In this study, the Weibull distribution with 𝛽 = 0.7, 0.9, 1.2, and 1.5 is considered, and at each 𝛽, the values of 𝛼 are
1
determined, using the expression 𝛼 = 𝜃∕Γ( + 1), for the desired TBE mean 𝜃. Table 12 shows the ARL and SDRL results
𝛽
ASLAM et al. 19

TA B L E 1 0 Run-length characteristics for competing control charts when 𝑘 = 3 and ARL0 ≅ 370
𝐓𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐆𝐖𝐌𝐀_𝐓𝐁𝐄 𝐆𝐖𝐌𝐀_𝐓𝐁𝐄
𝒒 = 𝟎.𝟖, 𝒒 = 𝟎.𝟗, 𝒒 = 𝟎.𝟖, 𝒒 = 𝟎.𝟗,
𝝀 =0.05 0.1 0.05 0.1 0.05 0.1 𝒂 = 0.7 𝒂 = 0.7 𝒂 = 0.7 𝒂 = 0.7
𝜹 𝑳 =1.285 1.599 1.498 1.793 1.993 2.112 1.954 1.553 2.059 2.033
1 370.55 370.00 370.64 370.37 370.54 370.29 370.87 369.90 370.41 370.30
(433.79) (387.18) (412.23) (380.95) (378.22) (372.99) (381.21) (437.88) (364.72) (363.05)
0.975 202.16 224.88 209.35 231.76 229.01 248.97 227.48 194.65 249.80 225.89
(231.17) (230.23) (227.89) (235.08) (232.80) (248.98) (231.22) (224.82) (241.24) (213.07)
0.95 122.91 144.31 129.44 152.32 148.62 172.59 147.89 115.48 172.93 147.04
(135.11) (142.97) (137.26) (150.75) (146.01) (168.65) (146.32) (130.16) (161.63) (129.44)
0.925 81.72 98.95 85.95 104.44 100.95 122.16 99.64 75.33 124.87 103.33
(86.63) (95.20) (88.08) (100.53) (96.52) (117.15) (94.48) (81.28) (113.18) (88.10)
0.9 57.45 70.32 60.17 74.71 72.12 88.78 71.55 52.43 92.42 75.38
(58.63) (64.85) (59.10) (69.51) (66.33) (83.00) (66.44) (53.98) (81.94) (59.95)
0.85 33.15 40.37 33.96 42.23 41.47 50.81 41.04 29.57 54.44 45.88
(31.80) (33.80) (30.97) (36.11) (34.75) (44.13) (34.71) (28.21) (43.92) (31.84)
0.8 21.51 26.32 21.86 26.92 26.82 32.34 26.55 18.80 35.41 31.42
(19.96) (20.03) (18.61) (20.76) (20.12) (25.55) (20.07) (16.67) (25.58) (18.99)
0.7 10.80 13.83 11.03 13.76 13.91 16.09 13.76 9.56 18.60 18.03
(9.65) (9.39) (8.54) (8.94) (8.66) (10.27) (8.67) (7.40) (10.55) (8.46)
0.5 3.87 5.49 4.25 5.56 5.83 6.53 5.67 3.94 8.14 8.76
(2.96) (3.12) (2.55) (2.71) (2.53) (2.75) (2.53) (2.13) (2.92) (2.62)
0.25 1.51 2.37 1.95 2.59 2.87 3.20 2.78 2.01 4.27 4.89
(0.71) (0.79) (0.66) (0.71) (0.70) (0.70) (0.69) (0.55) (0.79) (0.76)

TA B L E 1 1 Overall performance comparison of HWMA_TBE versus competing control chart when k = 1, 2, 3, λ = 0.05, 0.1 and
ARL0 ≅ 370
𝝀 = 𝟎.𝟎𝟓
𝒌=𝟏 𝒌=𝟐 𝒌=𝟑
Control Chart 𝐄𝐐𝐋 𝐑𝐀𝐑𝐋 𝐏𝐂𝐈 𝐄𝐐𝐋 𝐑𝐀𝐑𝐋 𝐏𝐂𝐈 𝐄𝐐𝐋 𝐑𝐀𝐑𝐋 𝐏𝐂𝐈
HWMA_TBE 16.0922 1.0000 1.0000 16.4142 1.0000 1.0000 15.9537 1.0000 1.0000
TEWMA_TBE 38.2449 2.3766 3.2560 29.4544 1.7944 2.0585 25.2745 1.5842 1.6457
DEWMA_TBE 39.3690 2.4465 3.3956 30.3500 1.8490 2.1783 26.0481 1.6327 1.7697
EWMA_TBE 43.8252 2.7234 4.1664 34.1905 2.0830 2.7640 29.2477 1.8333 2.2831
DGWMA_TBE 43.2629 2.6884 4.0966 33.7990 2.0591 2.7138 29.0567 1.8213 2.2423
GWMA_TBE 48.0747 2.9875 5.0120 38.9419 2.3724 3.5908 33.9620 2.1288 3.0811
𝝀 = 𝟎.𝟏
𝒌=𝟏 𝒌=𝟐 𝒌=𝟑
Control Chart 𝐄𝐐𝐋 𝐑𝐀𝐑𝐋 𝐏𝐂𝐈 𝐄𝐐𝐋 𝐑𝐀𝐑𝐋 𝐏𝐂𝐈 𝐄𝐐𝐋 𝐑𝐀𝐑𝐋 𝐏𝐂𝐈
HWMA_TBE 23.8180 1.0000 1.0000 26.1034 1.0000 1.0000 26.3317 1.0000 1.0000
TEWMA_TBE 68.8736 2.8917 2.9417 55.4298 2.1235 2.0983 47.9346 1.8204 1.7831
DEWMA_TBE 71.2517 2.9915 3.0438 57.3706 2.1978 2.1684 49.6118 1.8841 1.8424
EWMA_TBE 77.3751 3.2486 3.3409 63.4634 2.4312 2.4200 55.2968 2.1000 2.0697
DGWMA_TBE 68.5798 2.8793 2.9343 55.7202 2.1346 2.1107 48.3000 1.8343 1.8001
GWMA_TBE 74.7201 3.1371 3.2552 63.3088 2.4253 2.4488 56.0872 2.1300 2.1369
20 ASLAM et al.

TA B L E 1 2 ARL and SDRL values of HWMA_TBE control chart under exponential and Weibull distributions
Mean Values of 𝐓𝐁𝐄 Observations
𝝀 𝑳 Distribution 1 0.975 0.95 0.925 0.9 0.85 0.8 0.7 0.5 0.25
0.03 0.661 Weibull (𝛽 = 0.7) 271.63 73.32 40.65 27.34 20.17 12.62 8.99 5.41 2.81 1.56
(2231.49) (395.97) (169.51) (102.40) (66.05) (34.61) (21.62) (10.40) (3.47) (1.16)
Weibull (𝛽 = 0.9) 341.21 89.87 47.36 30.31 22.33 13.83 9.51 5.64 2.86 1.56
(2073.23) (342.55) (151.01) (84.17) (57.26) (30.51) (18.02) (8.58) (2.87) (1.04)
Exponential 370.44 96.06 50.18 32.04 23.49 14.16 9.80 5.70 2.88 1.56
(1961.20) (326.02) (143.73) (80.17) (54.03) (28.03) (17.26) (7.90) (2.66) (1.00)
Weibull (𝛽 = 1.2) 459.38 109.69 56.65 36.19 25.37 15.13 10.29 5.89 2.92 1.53
(1954.21) (310.10) (134.80) (76.67) (50.19) (25.51) (15.49) (7.09) (2.34) (0.93)
Weibull (𝛽 = 1.5) 583.22 133.43 64.57 40.42 28.08 16.03 10.68 6.01 2.95 1.49
(1881.36) (296.87) (123.48) (70.29) (45.14) (22.74) (13.45) (6.05) (2.03) (0.87)
0.05 0.775 Weibull (𝛽 = 0.7) 270.57 80.88 46.43 31.36 23.48 14.76 10.54 6.44 3.28 1.80
(1997.57) (387.17) (174.75) (103.47) (68.30) (36.76) (22.77) (11.32) (3.85) (1.32)
Weibull (𝛽 = 0.9) 341.21 100.21 53.72 36.00 26.48 16.56 11.58 6.92 3.42 1.85
(1830.12) (341.28) (150.67) (89.39) (59.31) (32.03) (19.71) (9.62) (3.23) (1.19)
Exponential 369.89 106.97 57.75 38.31 27.88 17.40 12.04 7.06 3.45 1.86
(1730.96) (324.43) (143.98) (84.25) (55.98) (30.44) (18.59) (8.98) (2.97) (1.14)
Weibull (𝛽 = 1.2) 457.34 122.95 66.18 43.33 30.89 18.55 12.78 7.27 3.54 1.89
(1715.07) (304.85) (137.40) (80.60) (52.17) (27.45) (17.03) (7.92) (2.60) (1.08)
Weibull (𝛽 = 1.5) 579.85 149.42 76.34 48.69 34.15 20.21 13.45 7.55 3.61 1.91
(1655.95) (289.97) (126.33) (72.65) (47.11) (24.87) (14.91) (6.87) (2.20) (1.02)
0.1 0.891 Weibull (𝛽 = 0.7) 271.81 84.65 48.94 34.06 25.26 16.36 12.00 7.46 3.85 2.13
(1929.72) (384.09) (173.68) (102.33) (67.37) (36.50) (23.10) (11.65) (4.03) (1.41)
Weibull (𝛽 = 0.9) 339.87 102.13 57.47 39.04 28.98 18.83 13.29 8.03 4.08 2.23
(1760.19) (333.82) (150.62) (86.65) (58.18) (32.33) (19.89) (9.77) (3.40) (1.22)
Exponential 370.44 110.37 61.27 41.57 30.70 19.57 13.80 8.27 4.14 2.28
(1673.90) (314.14) (140.16) (82.52) (55.68) (29.94) (18.80) (9.17) (3.14) (1.16)
Weibull (𝛽 = 1.2) 456.22 126.32 69.99 46.50 33.73 21.04 14.71 8.63 4.23 2.34
(1667.62) (296.13) (133.60) (77.69) (50.85) (27.52) (16.92) (8.14) (2.72) (1.04)
Weibull (𝛽 = 1.5) 590.61 154.59 81.05 52.42 37.69 22.89 15.81 8.99 4.29 2.43
(1657.57) (284.49) (122.97) (70.23) (46.29) (24.53) (15.16) (7.11) (2.26) (0.92)
0.2 0.954 Weibull (𝛽 = 0.7) 269.49 80.64 46.69 32.19 24.49 16.01 11.66 7.48 4.03 2.30
(2002.85) (380.81) (166.52) (98.06) (64.52) (35.08) (21.73) (10.84) (3.83) (1.30)
Weibull (𝛽 = 0.9) 338.22 98.74 53.99 36.93 27.79 18.10 13.02 8.17 4.26 2.40
(1847.03) (329.81) (142.66) (83.79) (55.54) (30.13) (18.57) (9.33) (3.24) (1.08)
Exponential 370.33 105.69 58.09 39.37 29.31 18.97 13.61 8.33 4.31 2.43
(1777.22) (312.94) (136.41) (79.04) (52.48) (28.45) (17.67) (8.62) (3.00) (0.99)
Weibull (𝛽 = 1.2) (478.65) (124.86) (67.15) (44.75) (32.64) (20.47) (14.59) (8.79) (4.40) (2.48)
1839.73 306.84 131.00 75.67 49.39 26.00 16.11 7.84 2.62 0.86
Weibull (𝛽 = 1.5) (687.02) (159.38) (80.98) (52.41) (37.50) (22.89) (15.88) (9.23) (4.44) (2.53)
2025.74 308.97 127.90 72.29 46.76 24.41 14.85 7.01 2.21 0.71

for the HWMA_TBE control chart when the TBE observations follow the Weibull distribution. The SDRL results are
provided in parenthesis. These results lead to the following conclusions.

(i) The ARL0 values are less than 370 when 𝛽 < 1; however, for 𝛽 > 1, these values are greater than 370 (see Table 12).
This happens because, for a given TBE mean, the variance of the Weibull distribution reduces as shape parameter
ASLAM et al. 21

FIGURE 2 RPC values for HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE under Weibull distribution with 𝛽 = 0.9

𝛽 rises. So, the TBE observations fluctuating with a smaller variation in an IC state, the probability of an IC TBE
observations being less than the lower control limit decreased, hence the ARL0 value increases. Similar conclusions
are also be observed in the studies of Alevizakos, et al.,12 and Pehlivan and Testik.38
(ii) The HWMA_TBE control chart is found to be more robust for a wider range of shifts when the shape parameter 𝛽 for
the Weibull distribution is close to 1. For instance, if 𝛽 = 0.9, the ARL1 values are relatively close to the anticipated
values when 𝛽 = 0.7, 1.2, or 1.5 (see Table 12).
(iii) The ARL1 values are departed from the anticipated values as the 𝜆 values decrease, that is, for a wider range of shifts,
the HWMA_TBE control chart is more robust for a smaller value of 𝜆. For instance, when the TBE mean is 0.9 and
TBE observations follow the Weibull distribution with 𝛽 = 0.9, the ARL1 value for the HWMA_TBE control chart
with (𝜆, 𝐿) = (0.03, 0.661) is 22.33, which is 5.19% less than the anticipated value of 23.49, whereas the ARL1 value
for the HWMA_TBE control chart with (𝜆, 𝐿) = (0.05, 0.775), (𝜆, 𝐿) = (0.1, 0.891), and (𝜆, 𝐿) = (0.2, 0.954) are 26.48,
28.98, and 27.79, which are 5.32%, 5.93%, and 5.46% smaller than the anticipated values of 27.88, 30.70, and 29.31,
respectively (see Table 12).

Furthermore, in order to investigate the robustness properties of the competing control charts, similar types of results
can be computed separately for each of the competing control chart.
In order to compare the robustness of the HWMA_TBE control chart to the robustness of the competing control charts,
the relative percentage change (RPC) measure can be used, which is defined as follows
( )
ARLW − ARLE
RPC = × 100%,
ARLE

where ARLW denotes the ARL values under Weibull distribution and ARLE refers to the ARL values under exponential
distribution. A control chart with RPC values close to zero is considered to be more robust. Figure 2 displays the RPC
values for the competing control chart under the assumption that the TBE data follow the Weibull distribution with shape
22 ASLAM et al.

parameter 𝛽 = 0.9. These results indicate that the ARL0 values for the HWMA_TBE control chart are closer to 370 as
compared to other control charts. Furthermore, for small-to-moderate shifts, that is, 0.8 ≤ 𝛿 < 1, the ARL1 values for the
HWMA_TBE control chart are more closer to the anticipated values.

6.2 Log-normal distribution

The log-normal distribution is denoted by LN(𝜇, 𝜎), where 𝜇 is known as location parameter, and 𝜎 is called as shape
parameter of the log-normal distribution. The log-normal distribution is defined by the pdf given as follows:
( ( )2 )
1 1 ln (𝑥) − 𝜇
𝑓 (𝑥; 𝜇, 𝜎) = √ exp − , 𝑥, 𝜎 > 0, −∞ < 𝜇 < ∞. (31)
𝑥𝜎 2𝜋 2 𝜎

In this study the log-normal distribution with 𝜎 = 0.94, 0.974, 1.092, 1.182 is considered, and the values of 𝜇 are calcu-
lated using 𝜎 values to obtain the desired TBE means 𝜃, that is, 𝜇 = ln(𝜃) − 0.5𝜎2 . Table 13 addresses the IC and OOC ARL
and SDRL (given in parenthesis) values for the HWMA_TBE control chart when the true distribution of the TBE obser-
vation is assumed to be log-normal distribution. Moreover, the ARL and SDRL values of the HWMA_TBE control chart
under exponential distribution are also provided for comparison purposes. The results in Table 13 includes the necessary
conclusions given as follows:

(i) In some cases, the ARL0 values are greater than 370, while for the others, they are less than 370 (see Table 13). This is
because, for a given TBE mean, the variance of a log-normal distribution increases as 𝜎 increases, and TBE fluctuating
with more variation in an IC state and the probability of an IC TBE being smaller than the control limit increases;
hence the ARL0 value reduces. Similar results can be seen in the study of Pehlivan and Testik.38
(ii) The HWMA_TBE control chart is more robust for a wider range of shifts with a smaller value of shape parameter
𝜎. For instance, with TBE mean less or equal to 0.9, if 𝜎 = 0.94 then the ARL1 values are the most nearer to the
anticipated than the ARL1 values if 𝜎 = 0.974, 1.092, and 1.182 (see Table 13).
(iii) For small to moderate shifts, that is, 𝛿 ≥ 0.7, the ARL1 values of the HWMA_TBE control chart higher than the
anticipated values as 𝜆 values increase. For example, when the TBE mean is equal to 0.95, and TBE observations
follow the log-normal with 𝜎 = 0.94, the ARL1 value for the HWMA_TBE control chart with (𝜆, 𝐿) = (0.03, 0.661) is
49.81, which is 0.74% less than the anticipated value of 50.18, while the ARL1 value for the HWMA_TBE control chart
with (𝜆, 𝐿) = (0.05, 0.775), (𝜆, 𝐿) = (0.1, 0.891), (𝜆, 𝐿) = (0.2, 0.954) are 60.45, 69.92, 78.37, which are 4.68%, 14.11%,
and 34.91% larger than the anticipated values of 57.75, 61.27, and 58.09 (see Table 13).

Additionally, Figure 3 shows the results of the RPC values for the competing control charts, whereas the TBE observa-
tions follow the LN(𝜎 = 0.94). These findings suggest that the ARL0 values for the HWMA_TBE chart are closer to 370
than the other control charts. Furthermore, for a wider range of shifts, the ARL1 values for the HWMA_TBE chart are
closer to the anticipated values as compared to the competing control charts.
Overall, the robustness study of the HWMA_TBE control chart indicates that the HWMA_TBE control chart is more
robust than the competing control charts for a smaller value of the smoothing parameter with moderate-to-large shifts
(𝛿 ≤ 0.8) in the process.

7 REAL-LIFE APPLICATIONS

The section offers two real-life applications for the practical implementation of the HWMA_TBE control chart. Subsec-
tion 7.1 provides the real-life dataset analysis that represents the failure time of the vertical boring machine, while Sub-
section 7.2 addresses the patient’s breast cancer survival data analysis.

7.1 Example 1

Boring machines are commonly used in the processes such as milling, drilling, boring, thread cutting, and face turning.
Boring machines can be horizontal or vertical, referring to the direction of the axis, and they use a spinning tool, known
ASLAM et al. 23

TA B L E 1 3 ARL and SDRL values of HWMA_TBE control chart under exponential and log-normal distribution
Mean Values of 𝐓𝐁𝐄 Observations
𝝀 𝑳 Distribution 1 0.975 0.95 0.925 0.9 0.85 0.8 0.7 0.5 0.25
0.03 0.661 Exponential 370.44 96.06 50.18 32.04 23.49 14.16 9.80 5.70 2.88 1.56
(1961.20) (326.02) (143.73) (80.17) (54.03) (28.03) (17.26) (7.90) (2.66) (1.00)
LN(𝜎 = 0.94) 399.57 95.60 49.81 31.48 22.98 13.55 9.38 5.51 2.75 1.48
(2557.32) (394.91) (173.29) (92.79) (62.95) (31.25) (18.88) (8.85) (2.93) (1.03)
LN(𝜎 = 0.974) 376.42 92.63 46.97 30.90 22.07 13.60 9.42 5.51 2.76 1.49
(2556.42) (405.45) (172.08) (99.38) (62.10) (33.19) (19.99) (9.28) (3.06) (1.06)
LN(𝜎 = 1.092) 308.35 80.14 43.22 28.31 20.37 12.55 9.05 5.36 2.73 1.50
(2575.72) (429.38) (188.73) (108.60) (69.07) (36.28) (22.33) (10.62) (3.52) (1.14)
LN(𝜎 = 1.182) 264.08 73.13 40.02 26.52 19.63 12.24 8.67 5.23 2.70 1.50
(2464.46) (457.08) (201.36) (114.08) (76.21) (40.19) (24.24) (11.60) (3.82) (1.21)
0.05 0.775 Exponential 369.89 106.97 57.75 38.31 27.88 17.40 12.04 7.06 3.45 1.86
(1730.96) (324.43) (143.98) (84.25) (55.98) (30.44) (18.59) (8.98) (2.97) (1.14)
LN(𝜎 = 0.94) 455.61 115.16 60.45 38.75 28.17 17.22 12.01 6.94 3.41 1.78
(2557.62) (427.91) (180.92) (100.69) (65.56) (34.66) (21.21) (9.81) (3.23) (1.19)
LN(𝜎 = 0.974) 424.28 110.56 58.99 37.71 27.30 16.86 11.73 6.88 3.39 1.76
(2532.44) (436.74) (186.09) (103.17) (67.35) (35.50) (21.73) (10.26) (3.41) (1.21)
LN(𝜎 = 1.092) 340.79 95.69 51.85 34.72 25.23 15.76 11.16 6.72 3.32 1.76
(2486.9) (460.79) (200.70) (117.82) (76.75) (39.28) (24.56) (11.83) (3.88) (1.32)
LN(𝜎 = 1.182) 291.77 87.88 48.05 32.29 23.28 15.11 10.83 6.40 3.26 1.74
(2462.86) (490.80) (214.86) (124.37) (80.56) (43.81) (27.69) (12.83) (4.29) (1.39)
0.1 0.891 Exponential 370.44 110.37 61.27 41.57 30.70 19.57 13.8 8.27 4.14 2.28
(1673.90) (314.14) (140.16) (82.52) (55.68) (29.94) (18.80) (9.17) (3.14) (1.16)
LN(𝜎 = 0.94) 587.12 133.13 69.92 44.65 32.73 20.21 14.23 8.39 4.21 2.26
(3228.68) (462.46) (193.96) (105.44) (70.09) (36.03) (22.30) (10.30) (3.43) (1.22)
LN(𝜎 = 0.974) 534.15 127.61 66.82 44.11 31.81 19.77 13.88 8.33 4.18 2.24
(3094.96) (469.05) (194.90) (110.05) (70.83) (36.89) (22.49) (10.90) (3.58) (1.25)
LN(𝜎 = 1.092) 408.99 111.40 60.31 39.65 29.00 18.60 13.14 8.02 4.08 2.18
(2830.53) (508.69) (216.87) (121.16) (77.79) (41.81) (25.23) (12.25) (4.12) (1.38)
LN(𝜎 = 1.182) 343.09 100.63 55.12 37.53 27.36 17.84 12.87 7.72 3.94 2.13
(2728.55) (537.37) (232.67) (136.87) (85.40) (46.42) (28.68) (13.40) (4.45) (1.48)
0.2 0.954 Exponential 370.33 105.69 58.09 39.37 29.31 18.97 13.61 8.33 4.31 2.43
(1777.22) (312.94) (136.41) (79.04) (52.48) (28.45) (17.67) (8.62) (3.00) (0.99)
LN(𝜎 = 0.94) 945.27 163.50 78.37 48.21 33.97 20.86 14.57 8.58 4.37 2.49
(5044.97) (607.99) (232.46) (119.81) (74.07) (37.84) (22.59) (10.36) (3.30) (1.03)
LN(𝜎 = 0.974) 805.28 151.34 73.68 46.05 33.41 20.36 14.24 8.51 4.37 2.48
(4586.67) (591.92) (226.70) (120.36) (77.63) (38.65) (23.08) (10.69) (3.46) (1.07)
LN(𝜎 = 1.092) 531.85 122.36 63.64 41.44 29.85 18.61 13.27 8.21 4.29 2.43
(3657.29) (578.15) (234.07) (129.71) (81.57) (40.85) (25.46) (12.06) (3.97) (1.23)
LN(𝜎 = 1.182) 412.56 108.88 58.96 37.31 27.91 17.76 12.87 7.97 4.19 2.39
(3223.44) (600.30) (255.71) (132.43) (87.17) (45.67) (28.1) (13.35) (4.38) (1.34)

as the machine spindle, to generate the primary motion. These machines generate smooth and precise hole diameters by
enlarging existing holes in the work piece. For this purpose, boring machines use a cutter, boring rod, drill, or milling
head to bore closed and opened in solid material, threading, reaming, boring, milling surfaces, and so on. The boring tool
may be a cemented carbide, a single steel cutting tip, a diamond, or a small grinding wheel. In a large-scale manufacturing
plant, special boring machines with several spindles are commonly used.39–42
24 ASLAM et al.

FIGURE 3 RPC values for HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE under log-normal distribution with 𝜎 = 0.94

TA B L E 14 Failure of vertical boring machine (in hours)


Sample Failure Sample Failure Sample Failure Sample Failure
number time number time number time number time
1 2802 9 2937 17 2136 25 4359
2 4020 10 1781 18 2816 26 2655
3 3886 11 2296 19 3158 27 3695
4 4155 12 3811 20 2380 28 376
5 2172 13 3705 21 2848 29 4339
6 2076 14 2672 22 3632 30 1976
7 1700 15 1596 23 1701 31 3575
8 3802 16 4351 24 4291 32 808

This example analyzes the dataset listed in Table 14, which addresses the failure times of the vertical boring machine.
Numerous authors have used this dataset in their studies. For instance, Majumdar43 used this dataset to design the optimal
management strategy for a vertical boring machine. Also, Krishna and Malik44 investigated several reliability models to
identify the best-fitted model for boring machine failures. Moreover, Hossain, et al.45 and Hossain, et al.46 implemented
this dataset in designing several control charts. The TBE control charts may be utilized for more than just monitoring the
processes of systems or components (Xie, et al.7 ); they are also used to process the boring machine failure. Every effort
should be made to maintain the boring process capability in order to minimize the cost and burden of boring machine
failure.
This dataset consists of 32 failure time (in hours) observations of the vertical boring machine. In order to imple-
ment the methodology of the HWMA_TBE control chart, the Kolmogorov-Smirnov (K-S) test is performed for the good-
ness of fit of the gamma distribution. The 𝑝-value for the K-S test is found to be 0.3053, which confirmed that the
failure time for the vertical boring machine followed the gamma distribution with parameters 𝑘 = 3 and 𝜃 = 959.3.
ASLAM et al. 25

TA B L E 1 5 Values of plotting statistics 𝑈𝑡 , 𝑊𝑡 , 𝑍𝑡 , and 𝑌𝑡 and the corresponding 𝐿𝐶𝐿𝑡 for the charts for boring machine failure dataset
𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐓𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐇𝐖𝐌𝐀_𝐓𝐁𝐄
𝒕 𝑿𝒕 𝒀𝒕 𝑳𝑪𝑳𝒕 𝒁𝒕 𝑳𝑪𝑳𝒕 𝑾𝒕 𝑳𝑪𝑳𝒕 𝑼𝒕 𝑳𝑪𝑳𝒕
1 2.1976 2.9599 2.8274 2.9980 2.9935 2.9999 2.9998 2.9599 2.9033
2 3.1529 2.9695 2.7619 2.9966 2.9861 2.9997 2.9995 2.2454 1.1595
3 3.0478 2.9735 2.7155 2.9954 2.9776 2.9995 2.9990 2.6939 1.6968
4 3.2588 2.9877 2.6793 2.9950 2.9684 2.9993 2.9982 2.8224 1.9344
5 1.7035 2.9235 2.6499 2.9915 2.9588 2.9989 2.9972 2.8538 2.0759
6 1.6282 2.8587 2.6252 2.9848 2.9490 2.9982 2.9958 2.6200 2.1724
7 1.3333 2.7825 2.6044 2.9747 2.9390 2.9970 2.9941 2.4399 2.2434
8 2.9820 2.7924 2.5864 2.9656 2.9291 2.9955 2.9922 2.3643 2.2986
9 2.3035 2.7680 2.5708 2.9557 2.9192 2.9935 2.9899 2.4076 2.3430
10 1.3969 2.6994 2.5573 2.9429 2.9094 2.9909 2.9874 2.3507 2.3798
11 1.8008 2.6545 2.5454 2.9285 2.8999 2.9878 2.9847 2.2755 2.4108
12 2.9890 2.6712 2.5349 2.9156 2.8905 2.9842 2.9817 2.2917 2.4375
13 2.9059 2.6830 2.5256 2.9040 2.8814 2.9802 2.9785 2.3457 2.4607
14 2.0957 2.6536 2.5174 2.8915 2.8726 2.9758 2.9751 2.3483 2.4811
15 1.2518 2.5835 2.5101 2.8761 2.8640 2.9708 2.9716 2.2880 2.4993
16 3.4125 2.6250 2.5037 2.8635 2.8558 2.9654 2.9679 2.3270 2.5157
17 1.6753 2.5775 2.4979 2.8492 2.8478 2.9596 2.9641 2.3080 2.5304
18 2.2086 2.5590 2.4927 2.8347 2.8402 2.9534 2.9601 2.2974 2.5439
19 2.4769 2.5549 2.4881 2.8207 2.8329 2.9467 2.9561 2.3059 2.5561
20 1.8667 2.5205 2.4840 2.8057 2.8258 2.9397 2.9520 2.2844 2.5674
21 2.2337 2.5062 2.4803 2.7907 2.8191 2.9322 2.9479 2.2819 2.5778
22 2.8486 2.5233 2.4770 2.7774 2.8127 2.9245 2.9437 2.3103 2.5874
23 1.3341 2.4638 2.4740 2.7617 2.8066 2.9163 2.9395 2.2591 2.5964
24 3.3655 2.5089 2.4713 2.7490 2.8008 2.9080 2.9352 2.3204 2.6047
25 3.4188 2.5544 2.4689 2.7393 2.7952 2.8995 2.9310 2.3666 2.6126
26 2.0824 2.5308 2.4668 2.7289 2.7899 2.8910 2.9268 2.3419 2.6199
27 2.8980 2.5492 2.4648 2.7199 2.7849 2.8825 2.9226 2.3727 2.6268
28 0.2949 2.4365 2.4631 2.7057 2.7802 2.8736 2.9184 2.2620 2.6333
29 3.4031 2.4848 2.4615 2.6947 2.7757 2.8647 2.9143 2.3471 2.6394
30 1.5498 2.4380 2.4601 2.6819 2.7714 2.8555 2.9102 2.2909 2.6453
31 2.8039 2.4563 2.4589 2.6706 2.7674 2.8463 2.9061 2.3289 2.6508
32 0.6337 2.3652 2.4577 2.6553 2.7635 2.8367 2.9022 2.2357 2.6560

Following the same lines as Alevizakos and Koukouvinos,13 the IC value 𝜃0 is assumed to be 1275, so that the decreas-
ing shift 𝛿 = 959.3∕1275 ≅ 0.75. Each failure time is rescaled by divided 1275 as the previous developments are assumed
𝜃0 = 1. For the predefined ARL0 = 370, the design parameters the HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and
EWMA_TBE control charts are chosen, respectively as, (𝜆, 𝐿) = (0.05, 1.117), (𝜆, 𝐿) = (0.05, 1.285), (𝜆, 𝐿) = (0.05, 1.498),
and (𝜆, 𝐿) = (0.05, 1.993) (see Table 1 and 10). Using these design parameters, respectively, the plotting statistics of the
HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts 𝑈𝑡 , 𝑊𝑡 , 𝑍𝑡 , and 𝑌𝑡 are computed along
with their corresponding lower control limit, 𝐿𝐶𝐿𝑡 , and presented in Table 15, with OOC points are shown in bold print.
The HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts are also depicted in Figure 4. The
results indicate that the HWMA_TBE control chart detects shift at the 10th failure number, while the TEWMA_TBE,
DEWMA_TBE, and EWMA_TBE control charts detect the OOC signal at the 15th, 18th, and 27th failure number, respec-
tively (see Table 17 and Figure 4). Moreover, the HWMA_TBE control chart detects overall 23 OOC points, while the
TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts detect 18, 15, and 7 OOC signals, respectively (see Table 17
26 ASLAM et al.

FIGURE 4 The HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts for Example 1 data

and Figure 4). This indicates that the HWMA_TBE control chart is more efficient than the TEWMA_TBE, DEWMA_TBE,
and EWMA_TBE control charts for small to moderate shifts in the process.

7.2 Example 2

The most invasive and rapidly spreading non-metastatic disease in women worldwide is breast cancer, and it is curable in
70% to 80% of patients in its early stages.47 According to the World Health Organization (WHO), 2.3 million women were
diagnosed with breast cancer in 2020, resulting in 685,000 deaths. Furthermore, 7.8 million women who survived at the
end of 2020 were diagnosed with breast cancer in the previous 5 years, making it the world’s most common disease. Breast
cancer is a global issue, and in the last five years, about 1.7 million new cases are diagnosed each year.48 Breast cancer
imposes a burden on early diagnosis, prevention, and treatment irrespective of the country’s economic status. Although
a lower incidence rate is seen in developing countries, the death rate is rising, suggesting that the developing countries
lack the resources for preventive screening, early diagnosis, and effective treatment.49 Breast cancer is a hazardous and
rapidly spreading disease that causes the death of millions, so every effort should be made to prevent, diagnose, and treat
it promptly in order to reduce the burden of disease and improve survival.
This example evaluates the dataset that illustrates the survival time for patients with breast cancer given by Awodutire,
et al.50 This dataset was collected from women with breast cancer who got treatment at the Ladoke Akinotola University
of Teaching Hospital in Osogbo, Nigeria, between 2000 and 2014. Awodutire, et al.50 fitted several models to this data;
one of them is an exponential distribution. In this example, the HWMA_TBE control chart along with TEWMA_TBE,
DEWMA_TBE, and EWMA_TBE control charts are implemented to the breast cancer survival data. The primary goal of
this application is to provide a monitoring scheme for the death rate of breast cancer patients in order to design feasible
ASLAM et al. 27

T A B L E 1 6 Survival time (in days) for breast cancer patients from 2000–2014 in Ladoke Akinotola University of Technology Teaching
Hospital Osogbo, Nigeria
Patient Survival Patient Survival Patient Survival Patient Survival
no. time no. time no. time no. time
1 48 14 50 27 1 40 151
2 23 15 21 28 48 41 250
3 297 16 35 29 16 42 139
4 151 17 287 30 59 43 10
5 10 18 6 31 127 44 23
6 7 19 50 32 24 45 13
7 48 20 151 33 17 46 7
8 8 21 10 34 219 47 168
9 2 22 45 35 251 48 315
10 1 23 300 36 6 49 312
11 98 24 3 37 161 50 60
12 229 25 1 38 329 51 198
13 3 26 308 39 242

screening plans, assist in the development of public health initiatives, or reduce the burden of breast cancer and improve
survival. If the mortality process goes OOC, the authorities must take corrective actions and need to update the planning,
screening, and preventive programs. Table 16 shows an uncensored dataset containing the survival times (in days) of 51
breast cancer patients.
In order to implement the HWMA_TBE control chart to this dataset, the rationale of the Alevizakos, et al.12 is fol-
lowed. It is supposed that the dataset is followed by the gamma distribution with parameters 𝑘 = 1 and 𝜃 = 88 and the
IC value 𝜃0 is assumed to be equal to 120. Because the previous developments assumed that 𝜃0 = 1, so each observa-
tion in the data set is divided by 120, such that the process shift 𝛿 = 88∕120 ≅ 0.73. The plotting statistics 𝑈𝑡 , 𝑊𝑡 , 𝑍𝑡 ,
and 𝑌𝑡 , for the HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts, respectively, are computed
along with their corresponding lower control limits 𝐿𝐶𝐿𝑡 and presented in Table 17, where the OOC are shown in bold
fonts. These statistics are computed on the basis of the design parameters (𝜆, 𝐿) = (0.05, 0.775), (𝜆, 𝐿) = (0.05, 1.247),
(𝜆, 𝐿) = (0.05, 1.460), and (𝜆, 𝐿) = (0.05, 1.868), for the HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE
control charts, respectively, at ARL0 ≅ 370.
The HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts are also displayed in Figure 5.
Table 17 and Figure 5 indicate that the HWMA_TBE control chart detects shift after patient number 2, whereas the
TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts detect the OOC signal after patient number 17, 14, and
15, respectively. Similarly, in overall detection ability, the HWMA_TBE control chart detects 43 OOC signals, while the
TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts detect 35, 28, and 2 OOC signals, respectively. These find-
ings suggested that the HWMA_TBE control chart outperforms the TEWMA_TBE, DEWMA_TBE, and EWMA_TBE
control charts in detecting small-to-moderate shifts in the process.

8 CONCLUSION, SUMMARY, AND RECOMMENDATION

In this study, a lower-sided homogeneously weighted moving average is proposed for monitoring the TBE that followed by
the gamma distribution. The proposed control chart is known as the HWMA_TBE control chart that detects the downward
shifts in the process. The Monte Carlo simulations are used for the computation of numerical results. At specific shifts, per-
formance indicators such as average run length (ARL) and standard deviation run length (SDRL) are computed to compare
the performance of the HWMA_TBE control chart with various competing TBE control charts. The competing TBE con-
trol charts include the EWMA (EWMA_TBE), double EWMA (DEWMA_TBE), triple EWMA (TEWMA_TBE), GWMA
(GWMA_TBE), and double GWMA (GEWMA_TBE) control charts. Similarly, to compare the overall performance of the
HWMA_TBE control chart, the performance indicators extra quadratic loss (EQL, relative average run length (RARL),
and performance comparison index (PCI) are computed. The HWMA_TBE control chart is compared to the one-sided
28 ASLAM et al.

T A B L E 1 7 Values of plotting statistics 𝑈𝑡 , 𝑊𝑡 , 𝑍𝑡 , and 𝑌𝑡 and the corresponding 𝐿𝐶𝐿𝑡 for the control charts with patient’s breast cancer
survival dataset
𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐓𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐇𝐖𝐌𝐀_𝐓𝐁𝐄
𝒕 𝑿𝒕 𝒀𝒕 𝑳𝑪𝑳𝒕 𝒁𝒕 𝑳𝑪𝑳𝒕 𝑾𝒕 𝑳𝑪𝑳𝒕 𝑼𝒕 𝑳𝑪𝑳𝒕
1 0.4000 0.9700 0.9066 0.9985 0.9964 0.9999 0.9998 0.9700 0.9613
2 0.1917 0.9311 0.8712 0.9951 0.9922 0.9997 0.9995 0.3896 0.2627
3 2.4750 1.0083 0.8460 0.9958 0.9874 0.9995 0.9990 0.4048 0.4780
4 1.2583 1.0208 0.8265 0.9970 0.9822 0.9994 0.9984 1.0340 0.5732
5 0.0833 0.9739 0.8105 0.9959 0.9768 0.9992 0.9975 1.0314 0.6298
6 0.0583 0.9281 0.7972 0.9925 0.9713 0.9989 0.9964 0.8405 0.6685
7 0.4000 0.9017 0.7859 0.9880 0.9657 0.9983 0.9952 0.7272 0.6969
8 0.0667 0.8600 0.7762 0.9816 0.9601 0.9975 0.9938 0.6638 0.7190
9 0.0167 0.8178 0.7678 0.9734 0.9545 0.9963 0.9922 0.5867 0.7368
10 0.0083 0.7773 0.7604 0.9636 0.9490 0.9946 0.9906 0.5229 0.7515
11 0.8167 0.7793 0.7540 0.9544 0.9437 0.9926 0.9887 0.5119 0.7640
12 1.9083 0.8358 0.7483 0.9484 0.9384 0.9904 0.9868 0.5942 0.7747
13 0.0250 0.7952 0.7433 0.9408 0.9333 0.9879 0.9847 0.6095 0.7840
14 0.4167 0.7763 0.7389 0.9325 0.9283 0.9852 0.9826 0.5841 0.7922
15 0.1750 0.7462 0.7349 0.9232 0.9235 0.9821 0.9803 0.5601 0.7994
16 0.2917 0.7235 0.7314 0.9132 0.9188 0.9786 0.9780 0.5403 0.8060
17 2.3917 0.8069 0.7283 0.9079 0.9144 0.9751 0.9756 0.6297 0.8119
18 0.0500 0.7691 0.7255 0.9010 0.9101 0.9714 0.9732 0.6163 0.8173
19 0.4167 0.7514 0.7230 0.8935 0.9060 0.9675 0.9707 0.6031 0.8222
20 1.2583 0.7768 0.7208 0.8877 0.9020 0.9635 0.9682 0.6354 0.8267
21 0.0833 0.7421 0.7188 0.8804 0.8982 0.9593 0.9657 0.6078 0.8309
22 0.3750 0.7238 0.7170 0.8726 0.8946 0.9550 0.9632 0.5974 0.8347
23 2.5000 0.8126 0.7154 0.8696 0.8912 0.9507 0.9606 0.6936 0.8383
24 0.0250 0.7732 0.7139 0.8647 0.8879 0.9464 0.9581 0.6484 0.8417
25 0.0083 0.7349 0.7126 0.8582 0.8848 0.9420 0.9556 0.6215 0.8448
26 2.5667 0.8265 0.7115 0.8567 0.8818 0.9378 0.9530 0.7249 0.8477
27 0.0083 0.7856 0.7104 0.8531 0.8790 0.9335 0.9505 0.6679 0.8505
28 0.4000 0.7663 0.7095 0.8488 0.8763 0.9293 0.9481 0.6630 0.8531
29 0.1333 0.7347 0.7086 0.8431 0.8738 0.9250 0.9456 0.6403 0.8556
30 0.4917 0.7225 0.7079 0.8370 0.8714 0.9206 0.9432 0.6407 0.8579
31 1.0583 0.7393 0.7072 0.8322 0.8691 0.9162 0.9408 0.6641 0.8601
32 0.2000 0.7124 0.7065 0.8262 0.8669 0.9117 0.9385 0.6339 0.8622
33 0.1417 0.6838 0.7060 0.8190 0.8649 0.9070 0.9362 0.6174 0.8642
34 1.8250 0.7409 0.7055 0.8151 0.8630 0.9024 0.9340 0.6872 0.8661
35 2.0917 0.8084 0.7050 0.8148 0.8612 0.8980 0.9318 0.7340 0.8679
36 0.0500 0.7705 0.7046 0.8126 0.8595 0.8938 0.9296 0.6707 0.8697
37 1.3417 0.7991 0.7043 0.8119 0.8579 0.8897 0.9275 0.7180 0.8713
38 2.7417 0.8962 0.7039 0.8161 0.8564 0.8860 0.9255 0.8049 0.8729
39 2.0167 0.9522 0.7036 0.8229 0.8549 0.8829 0.9235 0.8196 0.8744
40 1.2583 0.9675 0.7034 0.8302 0.8536 0.8802 0.9216 0.8124 0.8759
41 2.0833 1.0233 0.7031 0.8398 0.8523 0.8782 0.9197 0.8648 0.8773
42 1.1583 1.0301 0.7029 0.8493 0.8512 0.8768 0.9179 0.8482 0.8787
43 0.0833 0.9827 0.7027 0.8560 0.8501 0.8757 0.9161 0.8019 0.8800
44 0.1917 0.9432 0.7025 0.8604 0.8490 0.8749 0.9144 0.7906 0.8812
(Continues)
ASLAM et al. 29

TA B L E 17 (Continued)
𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐃𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐓𝐄𝐖𝐌𝐀_𝐓𝐁𝐄 𝐇𝐖𝐌𝐀_𝐓𝐁𝐄
𝒕 𝑿𝒕 𝒀𝒕 𝑳𝑪𝑳𝒕 𝒁𝒕 𝑳𝑪𝑳𝒕 𝑾𝒕 𝑳𝑪𝑳𝒕 𝑼𝒕 𝑳𝑪𝑳𝒕
45 0.1083 0.9014 0.7024 0.8624 0.8480 0.8743 0.9128 0.7728 0.8824
46 0.0583 0.8593 0.7022 0.8623 0.8471 0.8737 0.9112 0.7555 0.8836
47 1.4000 0.8863 0.7021 0.8635 0.8463 0.8732 0.9096 0.8075 0.8847
48 2.6250 0.9732 0.7020 0.8689 0.8455 0.8730 0.9081 0.8813 0.8858
49 2.6000 1.0546 0.7019 0.8782 0.8447 0.8733 0.9067 0.9164 0.8869
50 0.5000 1.0269 0.7018 0.8857 0.8440 0.8739 0.9053 0.8457 0.8879
51 1.6500 1.0580 0.7017 0.8943 0.8434 0.8749 0.9040 0.8963 0.8889

FIGURE 5 The HWMA_TBE, TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts for Example 2 data

competing TBE control charts based on the aforementioned performance indicators. The comparison indicates that the
one-sided HWMA_TBE control chart has superior performance than the competing TBE control charts.
Similarly, the HWMA_TBE control chart’s steady-state ARLs are investigated and compared with the zero-state ARLs.
The HWMA_TBE control chart’s steady-state ARLs are found to be larger than the zero-state ARLs. This happened
because the HWMA_TBE control chart used the time-varying control limits, that are narrow at the starting phase of
the process. Likewise, the Phase-II monitoring of the HWMA_TBE control chart is also investigated to study the impact
of Phase-I estimation for the scale parameter of the gamma distribution on the performance of the Phase-II HWMA_TBE
control chart. The findings show that estimating the unknown scale parameter of the gamma distribution from the Phase-
I IC sample has certain effect on the performance of the Phase-II HWMA_TBE control chart. In addition, the robustness
study for the HWMA_TBE control chart is performed when the TBE data follow a Weibull or log-normal distribution. The
30 ASLAM et al.

study demonstrates that the HWMA_TBE control chart is more robust for moderate to large shifts when the smoothing
parameter is small. The results also indicate that the HWMA_TBE control chart is more robust than the TEWMA_TBE,
DEWMA_TBE, and EWMA_TBE control charts. Finally, two real-life datasets are also analyzed for the practical imple-
ment the HWMA_TBE control chart. One real-life data are the boring machine failure data, while the other is the patient’s
breast cancer survival data. The real-life data analyses also show the superiority of the HWMA_TBE control chart over
the TEWMA_TBE, DEWMA_TBE, and EWMA_TBE control charts.
It should be noted that several concerns may require further investigation. For example, the two-sided HWMA_TBE
control chart can be designed to monitor the TBE and detect both upward and downward shifts in the process. The
HWMA_TBE control chart methodology can also be used in situations where the TBE observations can be modeled by
the Weibull or log-normal distribution.

D A T A AVA I L A B I L I T Y S T A T E M E N T
All the data available for this study are included in the manuscript.

ORCID
Majid Khan https://orcid.org/0000-0002-8654-373X
Syed Masroor Anwar https://orcid.org/0000-0001-7184-9579

REFERENCES
1. Shewhart WA. Economic Control of Quality of Manufactured Product. New York: D. Van Nostrand Company; 1931.
2. Gan FF. Designs of one- and two-sided exponential EWMA charts. J Qual Technol. 1998;30(1):55–69.
3. Lucas JM. Counted data CUSUM’s. Technometrics. 1985;27(2):129–144.
4. Vardeman S, Ray DO. Average run lengths for CUSUM schemes when observations are exponentially distributed. Technometrics.
1985;27(2):145–150.
5. Gan FF. Design of optimal exponential CUSUM control charts. J Qual Technol. 1994;26(2):109–124.
6. Chan LY, Xie M, Goh TN. Cumulative quantity control charts for monitoring production processes. Int J Prod Res. 2000;38(2):397–408.
7. Xie M, Goh TN, Ranjan P. Some effective control chart procedures for reliability monitoring. Reliabil Eng Syst Safe. 2002;77(2):143–150.
8. Zhang CW, Xie M, Liu JY, Goh TN. A control chart for the Gamma distribution as a model of time between events. Int J Prod Res.
2007;45(23):5649–5666.
9. Santiago E, Smith J. Control charts based on the exponential distribution: adapting runs rules for the t Chart. Qual Eng. 2013;25(2):85–96.
10. Rakitzis AC. Monitoring exponential data using two-sided control charts with runs rules. J Statist Comput Simulation. 2015;86(1):149–159.
11. Chakraborty N, Human SW, Balakrishnan N. A generally weighted moving average chart for time between events. Commun Stat Simul
Comput. 2017;46(10):7790–7817.
12. Alevizakos V, Koukouvinos C, Lappa A. Monitoring of time between events with a double generally weighted moving average control
chart. Qual Reliab Eng Int. 2019;35(2):685–710.
13. Alevizakos V, Koukouvinos C. A double exponentially weighted moving average chart for time between events. Commun Stat Simul Com-
put. 2020;49(10):2765–2784.
14. Alevizakos V, Chatterjee K, Koukouvinos C. A triple exponentially weighted moving average control chart for monitoring time between
events. Qual Reliab Eng Int. 2021;37(3):1059–1079.
15. Liu JY, Xie M, Goh TN, Sharma PR. A comparative study of exponential time between events charts. Qual Technol Quant Manag.
2006;3(3):347–359.
16. Shafae MS, Dickinson RM, Woodall WH, Camelio JA. Cumulative sum control charts for monitoring Weibull-distributed time between
events. Qual Reliab Eng Int. 2015;31(5):839–849.
17. Yang J, Yu H, Cheng Y, Xie M. Design of gamma charts based on average time to signal. Qual Reliab Eng Int. 2016;32(3):1041–1058.
18. Qu L, Khoo MBC, Castagliola P, He Z. Exponential cumulative sums chart for detecting shifts in time-between-events. Int J Prod Res.
2017;56(10):3683–3698.
19. Kumar N, Baranwal A. Design and implementation of qth quantile-unbiased tr-chart for monitoring times between events. Qual Reliab
Eng Int. 2019;35(4):1061–1080.
20. Alevizakos V, Koukouvinos C. Monitoring reliability for a gamma distribution with a double progressive mean control chart. Qual Reliab
Eng Int. 2021;37(1):199–218.
21. Alevizakos V, Koukouvinos C. A progressive mean control chart for monitoring time between events. Qual Reliab Eng Int. 2020;36(1):161–
186.
22. Hu X, Castagliola P, Zhong J, Tang A, Qiao Y. On the performance of the adaptive EWMA chart for monitoring time between events. J
Statist Comput Simulation. 2021;91(6):1175–1211.
23. Hunter JS. The exponentially weighted moving average. J Qua Technol. 1986;18(1):203–210.
24. Abbas N. Homogeneously weighted moving average control chart with an application in substrate manufacturing process. Comput Ind
Eng. 2018;120:460–470.
ASLAM et al. 31

25. Adegoke NA, Abbasi SA, Smith ANH, Anderson MJ, Pawley MDM. A multivariate homogeneously weighted moving average control chart.
IEEE Access. 2019;7:9586–9597.
26. Riaz M, Abbasi SA, Abid M, Hamzat AK. A new HWMA dispersion control chart with an application to wind farm data. Mathematics.
2020;8(12):2136.
27. Riaz M, Abbas Z, Nazir HZ, Abid M. On the development of triple homogeneously weighted moving average control chart. Symmetry.
2021;13(2):360.
28. Alevizakos V, Chatterjee K, Koukouvinos C. The extended homogeneously weighted moving average control chart. Qual Reliab Eng Int.
2021;37(8):2134–2155.
29. Zhang CW, Xie M, Goh TN. Economic design of exponential charts for time between events monitoring. Int J Prod Res. 2005;43(23):5019–
5032.
30. Balakrishnan N, Paroissin C, Turlot JC. One-sided control charts based on precedence and weighted precedence statistics. Qual Reliab Eng
Int. 2015;31(1):113–134.
31. Maravelakis PE, Castagliola P. An EWMA chart for monitoring the process standard deviation when parameters are estimated. Comput
Stat Data Anal. 2009;53(7):2653–2664.
32. Montgomery DC. Introduction to Statistical Quality Control. 7th ed.. John Wiley & Sons; 2013.
33. Awais M, Haq A. A new cumulative sum control chart for monitoring the process mean using varied L ranked set sampling. J Ind Prod
Eng. 2018;35(2):74–90.
34. Borror CM, Keats JB, Montgomery DC. Robustness of the time between events CUSUM. Int J Prod Res. 2003;41(15):3435–3444.
35. Liu JY, Xie M, Goh TN, Chan LY. A study of EWMA chart with transformed exponential data. Int J Prod Res. 2007;45(3):743–763.
36. Zhang M, Megahed FM, Woodall WH. Exponential CUSUM charts with estimated control limits. Qual Reliab Eng Int. 2014;30(2):275–286.
37. Hu X, Castagliola P, Zhong J, Tang A, Qiao Y. On the performance of the adaptive EWMA chart for monitoring time between events. J
Statist Comput Simulation. 2020:1–37.
38. Pehlivan C, Testik MC. Impact of model misspecification on the exponential EWMA charts: a robustness study when the time-between-
events are not exponential. Qual Reliab Eng Int. 2010;26(2):177–190.
39. Stephenson DA, Agapiou JS. Metal Cutting Theory and Practice. CRC Press; 2018.
40. Rao PN. Manufacturing Technology: Metal Cutting and Machine Tools. McGraw-Hill Education; 2000.
41. Knight WA, Boothroyd G. Fundamentals of Metal Machining and Machine Tools. CRC Press; 2019.
42. Youssef HA, El-Hofy H. Machining Technology: Machine Tools and Operations. CRC Press; 2008.
43. Majumdar SK. An optimum maintenance strategy for a vertical boring machine system. Operat Res Soc India. 1993;30:344–365.
44. Krishna H, Malik M. Reliability estimation in Maxwell distribution with progressively Type-II censored data. J Statist Comput Simulation.
2012;82(4):623–641.
45. Hossain MP, Omar MH, Riaz M. New V control chart for the Maxwell distribution. J Statist Comput Simulation. 2016;87(3):594–606.
46. Hossain MP, Sanusi RA, Omar MH, Riaz M. On designing Maxwell CUSUM control chart: an efficient way to monitor failure rates in
boring processes. Int J Adv Manufact Technol. 2019;100(5–8):1923–1930.
47. Harbeck N, Penault-Llorca F, Cortes J, et al. Breast cancer. Nat Rev Dis Primers. 2019;5:1–31.
48. da Costa, Vieira RA, Biller G, Uemura G, Ruiz CA, Curado MP, Breast cancer screening in developing countries. Clinics (Sao Paulo, Brazil).
2017;72(4):244–253.
49. Francies FZ, Hull R, Khanyile R, Dlamini Z. Breast cancer in low-middle income countries: abnormality in splicing and lack of targeted
treatment options. Am J Cancer Res. 2020;10(5):1568–1591.
50. Awodutire PO, Kolawole OA, Ilori OR. Data on the survival times of breast cancer patients in a Teaching Hospital, Osogbo. Data Brief.
2020;32:106109.

AU T H O R B I O G R A P H I E S

Muhammad Aslam is a professor of Statistics in the Department of Mathematics and Statistics, Riphah International
University, Pakistan. He also served as a professor in the Department of Statistics, Quaid-e-Azam University, Pakistan,
from 2003 to 2014. He obtained his Ph.D. in Statistics from the University of Wales, UK, in 1996. His current research
interests include Bayesian inference, statistical process control, and regression analysis.

Majid Khan is an assistant professor of Statistics, Government Post Graduate College, Haripur, Pakistan. He is cur-
rently a Ph.D. candidate in Statistics at the Department of Mathematics and Statistics, Riphah International Univer-
sity, Pakistan. His research interests include statistical process control, nonparametric techniques, and experimental
designs.

Syed Masroor Anwar is a faculty member in the Department of Statistics, University of Azad Jammu and Kashmir,
Pakistan. He recived his Ph.D. in Statistics from the Department of Mathematics and Statistics, Riphah International
32 ASLAM et al.

University, Pakistan. His research interests include statistical process control, Bayesian inference, applied statistics,
and reliability engineering.

Babar Zaman is a faculty member in the Department of Mathematics, University of Hafr Al batin, Hafr Al batin, Saudi
Arabia. He received his Ph.D. in Statistics from the Department of Mathematical Sciences, Faculty of Sciences, Uni-
versiti Teknologi Malaysia, Skudai, Malaysia. His research interests include statistical process control, artificial neural
networks, machine learning, deep learning, biostatistics, clinical data management, web-based diseases registries, and
integrated stochastic fuzzy method for sustainability assessment.

How to cite this article: Aslam M, Khan M, Anwar SM, Zaman B. A homogeneously weighted moving average
control chart for monitoring time between events. Qual Reliab Eng Int. 2021;1–32. https://doi.org/10.1002/qre.3032

You might also like