applied

sciences
Article
New Approach for Process Capability Analysis Using
Multivariate Quality Characteristics
Moath Alatefi 1,2, * , Abdulrahman M. Al-Ahmari 1,2 and Abdullah Yahia AlFaify 1,2

1 Industrial Engineering Department, College of Engineering, King Saud University,

P.O. Box 800, Riyadh 11421, Saudi Arabia; [email protected] (A.M.A.-A.); [email protected] (A.Y.A.)
2 Raytheon Chair for Systems Engineering, King Saud University, P.O. Box 800, Riyadh 11421, Saudi Arabia
* Correspondence: [email protected]

Abstract: The evaluation of manufacturing processes aims to ensure that the processes meet the
desired requirements. Therefore, process capability indexes are used to measure the capability of
a process to meet customer requirements and/or engineering specifications. However, most of
the manufacturing products have more than one quality characteristic (QC), in which case, the
multivariate QCs should be evaluated together using a single capability index. The research in this
article proposes a methodology for estimating the multivariate process capability index (PCI). First,
the dimensions of the multivariate QCs are reduced into a new single variable using the proportion
of the process specification region, by comparing each variable datapoint to its specification limits.
Moreover, nonnormal data are transformed to normality using a root transformation algorithm. Then,
a large data sample is generated using the parameters of the new variable. The generated data are
compared to the specification limits to estimate the percent of nonconforming (PNC). Finally, the
capability index of a given process datapoints is estimated using the PNC. Accordingly, managerial
insights for the implementation of the proposed methodology in real industry are presented. The
methodology was assessed by well-known multivariate samples from four different distributions, in
which an algorithm was developed for generating these samples with their given correlations. The
results show the effectiveness of the proposed methodology for estimating multivariate PCIs. Also,
the results from this research outperform the previous published results in most cases.

Citation: Alatefi, M.; Al-Ahmari, Keywords: process capability analysis; multivariate quality characteristics; nonnormal data
A.M.; AlFaify, A.Y. New Approach
for Process Capability Analysis Using
Multivariate Quality Characteristics.
Appl. Sci. 2023, 13, 11616. https:// 1. Introduction
doi.org/10.3390/app132111616
The variations of the manufacturing process have been investigated widely in order
Academic Editor: Konrad to evaluate the capability of the process to meet the desired specifications. In this regard,
Kulakowski various techniques have been used, including design of experiments and statistical process
Received: 9 September 2023
control. Process capability analysis has been particularly helpful in this regard [1,2]. As
Revised: 12 October 2023 industrial systems have advanced, process engineers have needed to thoroughly analyze
Accepted: 21 October 2023 and manage every aspect of their processes [3]. By using process capability analysis, we
Published: 24 October 2023 can assess manufacturing processes and make use of the resulting information to enhance
the capabilities of the processes under investigation to meet the required specifications.
It is widely accepted that the majority of manufactured goods possess multiple quality
characteristics (QCs) that are functionally correlated, indicating that they should be evalu-
Copyright: © 2023 by the authors. ated simultaneously. As a result, the assessment of product quality becomes more intricate
Licensee MDPI, Basel, Switzerland.
with an increase in the number of QCs. Therefore, there is a need for identifying capability
This article is an open access article
indices that can address the capabilities of nonnormal multivariate processes. For instance,
distributed under the terms and
Wang [4] conducted a process capability analysis for seven QCs of semiconductor products.
conditions of the Creative Commons
There is a significant amount of literature on the topic of multivariate process capability
Attribution (CC BY) license (https://
analysis, with several studies proposing different approaches. For example, Taam et al. [5]
creativecommons.org/licenses/by/
4.0/).
introduced the first multivariate capability index that employs process regions (PRs) and

Appl. Sci. 2023, 13, 11616. https://doi.org/10.3390/app132111616 https://www.mdpi.com/journal/applsci

Appl. Sci. 2023, 13, 11616 2 of 14

specification regions (SRs). Chen [6] developed the initial multivariate Cp that utilizes the
proportion of nonconformance (PNC). Additionally, Shahriari et al. [7] conducted a process
capability analysis to assess the effectiveness of multivariate quality characteristics (QCs).
Braun [8] examined the forms of process regions (PRs) and specification regions (SRs) to
create a novel process capability index (PCI). Castagliola et al. [9] analyzed bivariate process
capabilities and produced two indices using the PNC. Bothe [10] developed a technique to
calculate a multivariate C_pk index. Wang et al. [2] introduced a new index derived from
C_p and C_pk using a principal component analysis (PCA) decomposition.
There are numerous other studies in the literature concerning multivariate capability
analysis [4,11–36]. Wang and Chen [37] were the first to utilize principal component
analysis (PCA) in process capability analysis. Das and Dwivedi [28] utilized the g and h
multivariate process capability indices for nonnormal data, and their proposed index has
been favorably compared to other indices in the literature. However, the calculation of
this index involves complex computations. Ciupke [22] suggested a multivariate process
capability index (PCI) that can be employed for both normal and nonnormal quality control
(QC) processes. To evaluate multivariate nonnormal processes, he proposes the use of
one-sided models to determine the Process Ratio (PR), which is then compared to the
Specification Ratio (SR). Pan, Li, and Shih [20] built on the work of Pan and Lee [32] to
introduce a multivariate nonnormal PCI. They estimated the original probability density
function using a weighted standard deviation. Castagliola [38] defined the conventional
PCIs (Cp and Cpk) based on the proportion of nonconforming (PNC).
Previous research in this field has primarily focused on PCIs for multivariate normal
data, as evidenced by studies [9,39,40]. However, there have been some endeavors to
expand the application of the proportion of nonconforming (PNC) to nonnormal data.
Abbasi and Niaki [41] introduced a multivariate nonnormal PCI by utilizing the PNC
concept. To normalize nonnormal data, they applied the root transformation method
and subsequently used Monte Carlo simulation to estimate the PNC. Ahmad et al. [33]
explored process capability analysis for multivariate nonnormal data using the proportion
of nonconforming (PNC). They utilized the covariance distance (CD) to decrease the
dimensionality of the multivariate Quality Control (QC) data. While many studies in
the literature address nonnormal PCIs [9,11,42–50], only a few studies concentrate on
multivariate nonnormal process capabilities. However, the number of these studies has
been growing annually [4,11,25,27,43,51–59].
The majority of current multivariate Process Capability Indices (PCIs) rely on normality
theory and assume that multivariate Quality Control (QC) data conform to a normal
distribution. However, in reality, most QC data do not follow a normal distribution.
Additionally, some proposed methods are only applicable in certain situations or are only
suitable for a limited number of QCs and distributions [60]. Furthermore, the complexity
of statistical calculations can hinder implementation [20]. Therefore, developing a robust
multivariate PCI remains a significant research opportunity.
The existing methods for estimating the capabilities of nonnormal multivariate pro-
cesses are limited, and therefore, exploring different approaches is a valuable avenue for
research. This study aims to develop a multivariate capability index that considers the
correlations among quality characteristics (QCs) and applies it to assess manufacturing
processes. The proposed methodology is a general framework for the practitioners to
evaluate their process based on multivariate QCs. Regardless of the distribution of the real
data, this methodology is effective in estimating a multivariate process capability index
(MPCI). The algorithm of this research includes checking normality using the skewness
measure and then transforming the data into normality using the same measure. Also,
it is worth noting that most products have more than one correlated QC, which should
be evaluated together. This study presents an effective methodology for estimating the
MPCI for real case data. The proposed methodology has been tested for its accuracy of
estimation. MPCI estimation using the proposed methodology in this research is better
than the algorithm from the literature in term of estimation accuracy, as presented in the
Appl. Sci. 2023, 13, 11616 3 of 14

discussion section. To calculate the multivariate capability index, the proposed method
initially reduces the multi-dimensional data to a single variable by comparing the process
data to the specifications and determining their relative values. Next, the proportion of
nonconforming (PNC) data is used to estimate the process capability index.
The methodology of this research is presented in detail in Section 2. Section 3 describes
the experiment procedures for applying the proposed method. The results of this research
are presented in Section 4. Section 5 provides more discussions about the results along with
comparing the results with another method from the literature. Finally, Section 6 provides
managerial insights for implementing the proposed method for estimating MPI, and the
research is concluded in Section 7.

2. Research Methodology
2.1. Proposed Methodology
The study follows a methodology that involves a series of steps to calculate an ap-
proximate PCI for multivariate nonnormal data, as illustrated in Table 1. The methodology
starts by transforming collected data (Xi ) to normal distribution using root transformation
method. The root transformation approach looks for a correct root (r) of right-skewed
nonnormal data in which, when the data are brought up to the power r ( X r ), the skewness
in the converted data distribution would be basically small. The bisection technique is used
to calculate r’s value in accordance with the idea that a function switches its skewness sign
as it goes through zero. The bisection approach can half interval between zero and one in
successive iterations and finally identifies the root by evaluating the function in the midst
of an interval and substituting whichever limit has the same sign of skewness.

Table 1. Research methodology steps.

Steps Descriptions Comments

Collect the sample (X) by specifying the process, the X consists of p QCs, and p ≥ 2. Each QC consists of n
Step 1
number of its QCs (p), and the sample size (n). sample size.
Compute transformed variable (Y i ) using appropriate
Step 2 Yi = Yi ( Xi ); i = 1, 2, . . . , p.
transformation method.
Transform and standardize the specifications limits by
Step 3 Compute USLz and LSLz .
the same parameters of step 2.
Find relative variables (r i ) by dividing each variable Yi
Step 4 ri = USLy
in Yi by corresponding specification limit.
r2 +r1 +...+rn
Step 5 Find the average of ri . r= n
r
Step 6 Compute pooled standard deviation of (Y i ). (n1 −1)∗S2 1 +(n2 −1)∗S2 2
SP = n1 + n2 −2

Generate large (N) sample normal distribution with

Step 7 N = Total # of generated vectors.
mean of r and SP .
NNC = # of data out of USL & LSL
Step 8 Estimate portion of nonconforming (PNC).
PNC = NNC N .

Step 9 Estimate PCI using PNC. ϕ−1 (0.5+0.5(1− PNC ))

Cp = 3 .

Additionally, the technique is applied to produce the transformed specification limits

(USLy ) by the same transformation equation, where Yi represents the transformed data
and p is the number of QCs. In order to increase the similarity between all multivariate
variables, the data of each variable are presented as a fraction of their specification limits.
The transformed variables are then divided by their corresponding specification limits to
Appl. Sci. 2023, 13, 11616 4 of 14

form new relative variables and specifications. Also, specification limits are divided by
themselves. Consequently, each variable or QC will have the same specification limits.

Yi
ri = (1)
USLy

where Yi is the transformed normal data of variable i, and USLy is its upper specification
limit. This will reveal the same specification limits for all variables, so then it could be
treated as single variable. The relative variables are then averaged to one variable r.

r2 + r1 + . . . + r p
r= (2)
p

where p is the number of multivariate variables, and r is the average of each vector of the
relative multivariate data.
The standard deviation of the multivariate data is used to generate a single variable
with a large sample size. This is called pooled standard deviation, SP of Yi , which is used to
generate a large sample from a normal distribution with the mean of r and SP . Generating
the large sample size is performed to estimate the percent of nonconforming (PNC) data
(fraction of data out of specification limit). PNC is then used:
s
(n1 − 1) ∗ S2 1 + (n2 − 1)∗S2 2 + . . . + n p − 1 ∗S2 p


SP = (3)
n1 + n2 + . . . + n p − 2

where p is the number of multivariate variables, n p is the sample size of variable p, and S2 p
is the variance of variable p. Then, based on the relative specification limits, the PNC of
the generated sample is estimated to be used to estimate the MPCI. Finally, the following
formula is used to estimate the process PCI for two sides limits:

ϕ−1 (0.5 + 0.5(1 − PNC ))

Cp = (4)
3

where C p is the process capability index, ϕ−1 is the inverse of the normal distribution, and
PNC is the percent of nonconforming in the simulated data. Meanwhile, the one-sided
process capability index is estimated by the following formula:

ϕ−1 (1 − PNC )
Cp = .
3
There are two main assumptions for the proposed methodology to be valid:
- First, the data should be normally distributed; thus, the proposed method consists of
checking the normality assumption and transforming the nonnormal data into normal
using the explained root transformation technique.
- Second, the proposed multivariate process capability index is specific for correlated
data. Consequently, unrelated multivariate variables could be individually investi-
gated using a univariate capability index.

2.2. Evaluating the Proposed Methodology

To examine the effective estimation of the actual process capability index, the proposed
method will be applied for theoretical distributions with a known actual process capability
index. During the evaluation of the proposed methodology process, it is essential to provide
multivariate nonnormal data with an arbitrary distribution and determined correlation.
An algorithm has been developed that produces nonnormal multivariate data with the
required correlation. Initially, random vectors of the multivariate normal distribution are
generated, using the zero-mean vector and the covariance matrix that includes the desired
correlation coefficients. Then, data are generated utilizing the inverse of the cumulative
Appl. Sci. 2023, 13, 11616 5 of 14

distribution function (CDF) for each nonnormal variable. The parameters of the desired
nonnormal variable are used in the inverse CDF at the values of the normal distribution’s
cumulative distribution function. The multivariate nonnormal data produced by these
steps would thus have a specific distribution and correlation coefficients. The output
correlation between the variables is, however, quite close to the required correlation and
appears to vary somewhat across runs. As a result, the algorithm will go through multiple
iterations until the tolerance between the intended and output correlation is reached.
Table 2 presents four different samples’ properties from three distributions. Also, the
distribution type, the number of variables, specification limits, and parameters for each
variable, correlation matrix, and actual PCI for each sample are presented. For example,
the first sample is generated using gamma distribution, and it has two variables. The first
variable has an upper specification limit equal to 13 and is generated using shape and scale
parameters of 1 and 2, respectively. The second variable of this sample is generated using
a shape parameter of 2 and scale parameter of 3, and its upper specification limit is 26.
Also, the correlation coefficient of the first sample is 0.49, indicating a moderately positive
correlation. Finally, this sample has an actual PCI of 0.89. The proposed method will be
applied to the four samples in Table 2. Consequently, the proposed method’s performance
is evaluated with different samples from different distributions.

Table 2. Actual PCI for known theoretical distributions [41].

Distribution Variable USL α β Correlation Actual PCI

gamma X1 13 1 2 1 0.49
0.89
X2 26 2 3 0.49 1
X1 130 5 7 1 −0.37 0.58
gamma X2 58 6 3 −0.37 1 −0.28 1.18
X3 150 2 8 0.58 −0.28 1
X1 0.99 2 5 1 0.79
Beta 1.12
X2 0.99 4 4 0.79 1
X1 7 2 2 1 0.28 0.58
Weibull X2 9 4 3 0.28 1 0.49 1.28
X3 10 6 6 0.58 0.49 1

3. Experiments
The experiment of this research starts by generating samples with multivariate data
with the illustrated properties in Table 2. These samples are used to evaluate the perfor-
mance of the proposed method. Each variable will be generated with specified parameters.
However, the variables in each sample must have the given correlation coefficients.
It is essential for the validation procedure to generate multivariate data using theo-
retical random distribution parameters and specific correlations. The inverse transform
method is used for this purpose. The inverse transform method is a statistical technique
used to generate random numbers from a given probability distribution. It uses the cu-
mulative distribution function (CDF) of the desired distribution to transform a uniform
random variable into a random variable with the desired distribution.
For this purpose, a computer algorithm is used to generate the multivariate data with
the desired correlation coefficients between variables (Figure 1). To begin, a set of random
vectors is generated from the multivariate normal distribution, with a defined mean vector
and covariance matrix. The number of vectors generated is equal to the desired sample
size, and each vector has dimensions of 1 by p, where p represents the number of variables
or quality characteristics. The mean vector is set to zero, and the desired correlation matrix
represents the variance–covariance matrix between all variables that need to be generated.
Next, for each variable Xi , where i ranges from 1 to p, an n-by-1 vector is generated
using the inverse of the desired cumulative distribution function (CDF). The inverse CDF
uses the parameters of the corresponding Xi at the values of the cumulative distribution
 Appl. Sci. 2023, 13, x 6 of 14

the parameters of the corresponding 𝑋 at the values of the cumulative distribution func-
tion of the normal distribution. Despite this, the output correlation between variables is
Appl. Sci. 2023, 13, 11616 almost equal to the desired correlation, and it varies in each run of the previous proce- 6 of 14
dures. Therefore, the algorithm is run for several iterations until the difference between
the desired and output correlation is less than suggested tolerance ε.
Furthermore, the algorithm for generating multivariate data with the desired corre-
function
lation is used toof the normal
generate distribution.
data from Despite
the four samples’ this, the
parameters in output correlation
Table 2. To examine between variables is
almost equal
the performance to suggested
of the the desired correlation,
multivariate anddifferent
PCI with it variessample
in each run
sizes, of dif-
four the previous procedures.
ferent sample sizes
Therefore, have
the been generated
algorithm is runfrom each sample.
for several The sample
iterations untilsizes
theare 50, 100, between the desired
difference
500, and 1000. For example, one sample of two variables from the gamma distribution
and output correlation is less than suggested tolerance ε.
with a sample size of 50 has been generated.

Figure 1. Generating multivariate data with desired correlation.

Figure 1. Generating multivariate data with desired correlation.
Then, the generated data (𝑋 ) is transformed to normal distribution (𝑌 ). This is done
by reducing Furthermore,
the skewness ofthe the algorithm
data. Since afor generating
normal multivariate
distribution data the
has zero skewness, with the desired correla-
tion of
skewness is the
used to generate
generated data from
data samples must be the four samples’
reduced. parameters
There are some common meth- in Table 2. To examine the
ods for reducing skewness
performance of theinsuggested
data such as logarithmic transformation
multivariate and square-root
PCI with different sample sizes, four different
transformation. In this research, the root transformation is used to reduce the skewness of
sample sizes have been generated from each sample. The sample sizes are 50, 100, 500, and
the data. Root transformation looks for an optimal root (r) such that if the data were raised
1000.
to the power Forr (𝑋example, one sample
), the skewness of twodata
in the resulting variables from
distribution the gamma
would be almostdistribution
zero. with a sample
sizepurpose,
For this of 50 has been code
a software generated.
was developed to search for the optimal root that would
reduce the Then,
skewness theof agenerated
given variable data
datapoint (𝑋 transformed
( X i ) is ). This code is based
to normal distribution (Y i ). This is
on the iterative
search
doneeither
byuntil the skewness
reducing is reducedof
the skewness to the
around
data.zero or after
Since performing
a normal a given
distribution has zero skewness,
number of iterations without improvement. Also, the specification limits of a given varia-
the skewness of the generated data samples must be reduced. There are some common
ble are transformed using the same root.
methods
The transformedfor reducing
variables skewness inthen
and limits are datadivided
such asby logarithmic transformation
their corresponding speci- and square-root
transformation. In this research, the root transformation is
fication limits to form new relative variables and specifications. The resulting specification used to reduce the skewness
limitsoffor all variables are the same. Therefore, the relative variables are averaged
the data. Root transformation looks for an optimal root (r) such that if the data were to form
a single variable. In this way, thermultivariate variables are dimensioned to univariate.
raised to the power r (X ), the skewness in the resulting data distribution would be almost
The mean and standard deviations of the averaged relative data are estimated to be used
zero. For
in estimating thethis purpose,
multivariate a software
process capabilitycode
index.was
Usingdeveloped
the mean and to standard
search for de- the optimal root that
would
viation, data reduce the skewness
are generated from the normalof a given variable
distribution with datapoint
a large sample( Xsize
i ) . This
(n = code is based on the
iterative search either until the skewness is reduced to around zero or after performing a
given number of iterations without improvement. Also, the specification limits of a given
variable are transformed using the same root.
The transformed variables and limits are then divided by their corresponding specifi-
cation limits to form new relative variables and specifications. The resulting specification
limits for all variables are the same. Therefore, the relative variables are averaged to form a
single variable. In this way, the multivariate variables are dimensioned to univariate. The
mean and standard deviations of the averaged relative data are estimated to be used in esti-
mating the multivariate process capability index. Using the mean and standard deviation,
data are generated from the normal distribution with a large sample size (n = 1,000,000).
Each single datapoint is compared with the relative specification in which nonconforming
data are counted. Then, the percentage of nonconforming (PNC) data is computed by
dividing the nonconforming data by the sample size (n). Finally, the PNC is utilized to
compute the PCI using Equation (4). Generating the normal data and computing the PCI is
 conforming data are counted. Then, the percentage of nonconforming (PNC) data is com-
puted by dividing the nonconforming data by the sample size (n). Finally, the PNC is
utilized to compute the PCI using Equation (4). Generating the normal data and compu-
ting
Appl. Sci. 2023, the PCI is performed 1000 times before the software displays the average PCI. Ten 7 of 14
13, 11616
replications are run to estimate the standard deviation of the estimated PCI.

4. Results performed 1000 times before the software displays the average PCI. Ten replications are
run to estimate the standard deviation of the estimated PCI.
The results of this research are related to three main subjects. These subjects are the
quality of generating multivariate data with a desired correlation, skewness reduction of
4. Results
the generated data, andThe theresults
effectiveness of theare
of this research proposed
related tomethod
three main insubjects.
estimatingThesethe actualare the
subjects
PCI. The generated data should meet both the given parameters and correlation coeffi-
quality of generating multivariate data with a desired correlation, skewness reduction of
the generated
cients. Figure 2 shows data, and the
the distribution effectiveness
fitting of the proposed
of the generated data. method
One in estimatingdata
variable’s the actual
PCI. The generated data should meet both the given parameters and correlation coefficients.
from each distribution was fitted to show the ability of the software code (Matlab) to gen-
Figure 2 shows the distribution fitting of the generated data. One variable’s data from
erate data with the each
desired distribution
distribution andtoparameters.
was fitted The
show the ability of upper rightcode
the software and(Matlab
left graphs
R2023a) to
of Figure 2 are for the samedata
generate variable
with thegenerated from the
desired distribution andgamma distribution
parameters. withand
The upper right shape
left graphs
and scale parameters of one and two, respectively. The upper left graph is the densityshape
of Figure 2 are for the same variable generated from the gamma distribution with
and scale parameters of one and two, respectively. The upper left graph is the density
function, whereas the graph in the upper right corner is the cumulative density function
function, whereas the graph in the upper right corner is the cumulative density function of
of the same variablethe
data.
sameAlso, onedata.
variable variable
Also, from both the
one variable from Beta
bothand Weibull
the Beta distributions
and Weibull distributions
were plotted using the
werecumulative
plotted usingdensity function.
the cumulative Allfunction.
density the fitted Alldata show
the fitted very
data showclose
very fit
close fit
and parameters. and parameters.

Figure 2. DistributionFigure
fitting2.of generated
Distribution data.
fitting of generated data.

Furthermore,coefficients
Furthermore, the correlation the correlation coefficients
between between generated
generated variablesvariables
in eachin each sample
sample
should be the same as the reference data in Table 2. The correlation between variables has a
should be the same significant
as the reference data in Table 2. The correlation between variables has
effect on the estimation of the PCI; that is, the greater the positive correlation, the
a significant effect on the
less estimation
variation there isofbetween
the PCI; that is,Consequently,
variables. the greater the positive correlation,
the correlation between generated
the less variation there is between
variables variables.
in each sample Consequently,
is presented in Table 3.the
Thecorrelation
results showsbetween gen-
that the correlations
between
erated variables in each the generated
sample variables
is presented in and
Tablethe3.actual
The correlations are very
results shows thatclose
the to each other.
corre-
lations between the generated variables and the actual correlations are very close to each
other. However, the data were generated in four different sample sizes (50, 100, 500, and
Appl. Sci. 2023, 13, 11616 8 of 14

Appl. Sci. 2023, 13, x 8 of 14

However, the data were generated in four different sample sizes (50, 100, 500, and 1000),
and the correlations of all different sample sizes are close to the actual correlation. It is
1000), and
noticed thatthe
thecorrelations
samples of of allvariables
two different have
sample sizes
more are close
accurate to the actual
correlations correlation.
than those of
It is noticed
three variables.that the samples of two variables have more accurate correlations than those
of three variables.
Table 3. Comparison of correlation between generated variables with actual.
Table 3. Comparison of correlation between generated variables with actual.
Distribution Gamma Gamma Beta Weibull
Distribution Gamma Gamma Beta Weibull
Variables X1, X2 X1, X2 X1, X3 X2, X3 X1, X2 X1, X2 X1, X3 X2, X3
Variables X1, X2 X1, X2 X1, X3 X2, X3 X1, X2 X1, X2 X1, X3 X2, X3
n = 50n = 50 0.491 0.491 −0.367 −0.3670.574 0.574−0.284−0.284 0.7900.790 0.288
0.288 0.592
0.592 0.494
0.494
Sample n = 100 0.490 −0.365 0.582 −0.275 0.790 0.286 0.576 0.494
n = 100 0.491 0.490−0.368 −0.3650.584 0.582−0.281−0.275 0.7900.790 0.286 0.576 0.494
size size n = 500
Sample 0.293 0.576 0.487
n = 500 0.491 0.491
n = 1000 −0.366 −0.368
−0.366 0.584−0.265−0.281 0.7920.790 0.293
0.293 0.576
0.580 0.487
0.494
Actual correlationn = 1000 0.49 0.491−0.37 −0.3660.58 −0.366−0.28−0.265 0.790.792 0,290.293 0.580
0.58 0.494
0.49
Actual correlation 0.49 −0.37 0.58 −0.28 0.79 0,29 0.58 0.49

Moreover,
Moreover,estimating
estimatingthethePCI
PCIrequired
requiredthethedata
datatotobebenormally
normallydistributed.
distributed.InInthis
this
regard, the results of applying root transformation for skewness reduction are
regard, the results of applying root transformation for skewness reduction are presented presented
here.
here.Figure
Figure33shows
showsthetheskewness
skewnessofofwhat
whatwaswasgenerated
generatedbefore
beforeand
andafter
afterapplying
applyingroot
root
transformation.
transformation. The results ensure the effectiveness of root transformation inreducing
The results ensure the effectiveness of root transformation in reducingthethe
skewness
skewnessofofthethedata.
data.Most
Mostofofthe
theskewness
skewnessof ofthe
thedata
dataafter
aftertransforming
transformingisisclose
closeto
tozero.
zero.
The
Themost
mostskewness
skewnessofofthethegenerated
generateddata
databefore
beforetransformation
transformationwas wasassociated
associatedwith
withthe
the
first variable of the gamma bivariate sample. However, after transforming the data using
first variable of the gamma bivariate sample. However, after transforming the data using
the root transformation method, the skewness was reduced to almost zero. It is worth
the root transformation method, the skewness was reduced to almost zero. It is worth
noting that these are samples of size 1000 data.
noting that these are samples of size 1000 data.

2.000

1.500

1.000

0.500

0.000

-0.500

Befor After

Figure3.3.Skewness
Figure Skewnessofofthe
thegenerated
generateddata
databefore
beforeand
andafter
afterroot
roottransformation.
transformation.

Theultimate
The ultimategoal
goalofofthis
thisresearch
researchisistotocome
comeup upwith
withaacapability
capabilityindex
indexthat
thatevaluates
evaluates
processcapability
process capabilitythrough
throughmultivariate
multivariatequality
qualitycharacteristics
characteristics(QCs).
(QCs).Multivariate
Multivariateprocess
process
capabilityanalysis
capability analysisisisconducted
conductedto toevaluate
evaluatethethecapability
capability of
ofaagiven
givenprocess,
process,which
which pro-
pro-
duces products with multivariate quality characteristics (QCs) to meet
duces products with multivariate quality characteristics (QCs) to meet customer and/or customer and/or
engineeringrequirements.
engineering requirements. In In this
thisresearch,
research,the theprocess
processperformance
performanceisismeasured
measuredusing
using
theprocess
the processcapability
capabilityindex,
index,ininwhich
whichaaprocess
processwith
withaaprocess
processcapability
capabilityindex
indexofofone
oneor
or
moreisisconsidered
more consideredcapable
capableofofproducing
producingproducts
productswith
withgiven
givenspecification
specificationlimits,
limits,whereas
whereas
aaprocess
processcapability
capabilityindex
indexofof less
less than
than oneone indicates
indicates thatthat
thethe process
process maymay produce
produce prod-
products
ucts outside
outside of theof the specification
specification limits.limits.
That That is because
is because the process
the process capability
capability indexindex divides
divides the
the difference between the mean of the process by the three times of standard deviation,
Appl. Sci. 2023, 13, 11616 9 of 14

difference between the mean of the process by the three times of standard deviation, in
which the three iterations of standard deviation are the normal tolerance for each process.
By applying the proposed methodology, it is supposed that the resulting PCI reflects the
true performance of the considered process. To measure the effectiveness of the proposed
multivariate process capability, Table 4 shows the results of multivariate process capability
indices for given distributions using the proposed method, along with their actual PCIs.
Samples from gamma, Beta, and Weibull distributions have been used in this research.
The results show that proposed methodology reveals multivariate PCIs close to the ac-
tual. The first sample consists of two variables, and the results of applying the proposed
methodology on this data revealed multivariate PCIs around 0.9, depending on the sample
size. The actual multivariate PCI of this sample is 0.89. It is noted that for this sample,
the greater the sample size is, the closer the proposed multivariate PCI is to the actual.
Also, the proposed multivariate PCIs for the other samples are close to the actual, either
underestimated or overestimated, such as the sample of gamma distribution with three
variables. However, the Beta distribution sample results are close to the actual multivariate
PCI but all underestimated from the actual for all sample sizes. Finally, the results of the
Weibull sample also show close results to the actual, especially for large samples. However,
all the results of the Weibull samples are overestimated.

Table 4. Results of proposed multivariate process capability indices.

CP
Distribution n USL α β Correlation
Mean Std Actual
50 0.932 0.000
gamma 100 13 1 2 1 0.49 0.895 0.000
0.89
500 26 2 3 0.49 1 0.872 0.006
1000 0.892 0.000
50 130 5 7 1 −0.37 0.58 1.155 0.001
gamma 100 58 6 3 −0.37 1 −0.28 1.2117 0.001
1.18
500 150 2 8 0.58 −0.28 1 1.1806 0.001
1000 1.2050 0.000
50 0.957 0.000
100 1 2 5 1 0.79 1.088 0.000
Beta 1.12
500 1 4 4 0.79 1 1.041 0.000
1000 1.102 0.001
50 7 2 2 1 0.28 0.58 1.685 0.010
100 9 4 3 0.28 1 0.49 1.125 0.000
Weibull 1.28
500 10 6 6 0.58 0.49 1 1.278 0.000
1000 1.284 0.000

5. Discussion
The results of this research emphasize the significance of the proposed methodology
in processing real data and estimating the process capability index of a given process
through multivariate QCs. Generating special data that follow a specific distribution with
specific parameters and a determined correlation is a very important research direction.
Transforming this method into a software code would facilitate studying the effect of
correlation in process capability analysis. This could be done by generating data from
the same distribution with the same parameters and different levels of correlation. Also,
combining the transformation method with generating data in one algorithm could be
utilized to search for samples that could be transformed to normality easily. The part of
skewness reduction in the algorithm could be combined in practice with other approaches
that may enhance the other properties of normal distribution, such as bringing the mean,
mode, and median close to each other.
Appl. Sci. 2023, 13, 11616 10 of 14

The presented results in the previous section show that the proposed methodology has
the ability to estimate multivariate PCIs precisely, since almost all the estimated multivariate
PCIs are very close to the actual. Moreover, the proposed methodology in this research has
been applied to bivariate samples and samples with three variables. Both types of samples
showed high performance.
Furthermore, the samples investigated in this research have been studied in the
literature using different methodologies. In this regard, Abbasi [41] estimated the
multivariate PCIs of the same 16 samples used in this research. They presented a good
estimation comparing to the actual PCIs. However, the results in this research outperform
their results in most cases. Table 5 presents a comparison between our research results
and results from previous research. Both studies revealed little variation from the actual
PCIs. These variations are due to either overestimating or underestimating the actual
PCIs. To present a clear comparison between the results from this research and the
results from previous research, the mean absolute percentage error (MAPE) is computed
for each sample for both approaches.

Table 5. Comparing research results with previous results.

Results [41]
Sample Actual
Cp MAPE Cp MAPE
1 0.932 4.7% 0.864 2.9% 0.89
2 0.895 0.6% 0.865 2.8% 0.89
3 0.872 2.0% 0.872 2.0% 0.89
4 0.892 0.2% 0.888 0.2% 0.89
5 1.155 2.1% 1.208 2.4% 1.18
6 1.2117 2.7% 1.258 6.6% 1.18
7 1.1806 0.1% 1.148 2.7% 1.18
8 1.205 2.1% 1.172 0.7% 1.18
9 0.957 14.6% 0.885 21.0% 1.12
10 1.088 2.9% 1 10.7% 1.12
11 1.041 7.1% 0.997 11.0% 1.12
12 1.102 1.6% 0.975 12.9% 1.12
13 1.685 31.6% 1.411 10.2% 1.28
14 1.125 12.1% 1.298 1.4% 1.28
15 1.278 0.2% 1.267 1.0% 1.28
16 1.284 0.3% 1.297 1.3% 1.28

Mean absolute percentage error (MAPE) is a measure of estimation accuracy in statis-

tics. It usually expresses the accuracy as a ratio defined by the formula:

| APCI − EPCI |
MAPE = ∗ 100% (5)
APCI
where APCI is the actual MPCI, and EPCI is the estimated MPCI. MAPE can be interpreted
as the average percentage error of the MPCI estimation compared to the actual values.
A lower MAPE indicates a better estimation and a higher accuracy. MAPE is commonly
used in model evaluation because of its very intuitive interpretation in terms of relative
error. According to the MAPE measure, the research estimation of PCIs outperforms the
previous results in 12 samples out of 16 samples. In particular, from eight samples of
gamma distribution, the research estimation perform better in six samples, whereas PCI
estimations of Beta distribution samples using this research approach are always better.
Regarding the Weibull distribution samples, the approach of this research performs well
with large samples.
The variations among the estimated multivariate PCIs are due to the associated ran-
domness in the data. The randomness sources exist in different steps of the proposed
methodology, starting with the data generation of the used data for evaluating the pro-
posed methodology or for estimating the PNCs. The variations observed in multivariate
Appl. Sci. 2023, 13, 11616 11 of 14

data generation using the Monte Carlo simulation from one run to another depend on
several factors, such as the number of samples drawn from the distribution. The larger
the number of samples, the more accurate and representative the synthetic data will be
of the true distribution. However, larger sample sizes also require more computational
resources and time. Another factor is the random number generator used to draw the
samples. The quality and reliability of the random number generator affect the randomness
and independence of the samples. A poor random number generator may introduce bias
or correlation in the synthetic data that is not present in the true distribution. Therefore,
generating data from a specific random distribution with specific parameters many times
will result in different data in each run.
Moreover, the proposed methodology involves randomness in terms of searching
for the best root transformation. Each run may result in a different root for transforming
the data and will generate different sample for comparing with the transformed specifi-
cation limits. Consequently, slight differences of multivariate PCIs are experienced from
different replications.

6. Managerial Insights
Multivariate process capability indices (MPCIs) are useful tools to evaluate the quality
and performance of a manufacturing process that involves two or more correlated product
characteristics. However, implementing MPCI in real-world contexts may face some
challenges and constraints, such as:
• Collecting and analyzing sufficient and representative data from the process. To
calculate MPCI, one needs to have enough data from the process to estimate the
parameters of the joint distribution and variation of the product characteristics. The
data should also be representative of the normal operating conditions of the process,
without any special causes of variation or outliers. Moreover, the data should be
collected in a timely and efficient manner, using appropriate sampling techniques and
measurement systems.
• Communicating and interpreting the results of MPCI to stakeholders. MPCIs are
numerical measures that quantify the capability of a multivariate process, but they
may not be easy to understand or communicate to stakeholders who are not familiar
with statistics or quality engineering. Therefore, it is important to present and interpret
the results of MPCI in a clear and meaningful way, using graphical displays, tables,
or verbal descriptions. For example, one can use scatterplots or contour plots to
visualize the joint distribution and variation of the product characteristics, as well
as the tolerance region and the process region. One can also use tables to compare
different MPCIs or different processes based on their values or rankings. One can
also use verbal descriptions to explain what MPCI mean in terms of proportion of
nonconforming units or customer satisfaction.
These are some insights into how the proposed methodology could be practically
implemented in manufacturing contexts, considering the real-world challenges and con-
straints. Furthermore, in the real industry, especially in manufacturing organizations, most
of their products have more than one quality characteristics. Therefore, the practitioners
need to evaluate their outputs using a single index that combine all the QCs of one product.
Actually, the QCs of one product are often functionally correlated with each other, in which
all QCs should be evaluated together. Consequently, this proposed methodology can help
quality professionals and practitioners to evaluate their processes. Moreover, the proposed
methodology suggests a way for transforming the nonnormal data into normality, as most
of the real case data are not normally distributed. Generally, the proposed methodology
in this research has provided a general framework for evaluating real case processes with
multivariate data.
Appl. Sci. 2023, 13, 11616 12 of 14

7. Conclusions
This research presents a general framework that leads to estimating process capability
indexes for multivariate data. Multivariate data exists in a wide range in real industry
applications. Most of the products have more than one characteristic that is critical to
quality from customer and engineering specifications viewpoints, which leads investigators
to collect multiple responses for the multi-quality characteristics. The collected multivariate
data should be analyzed together due to their functional correlation. The presented frame-
work in this paper provides solutions for most issues associated with multivariate data,
including the normality assumption, dimension reduction, and performance evaluation
using multivariate PCIs. Moreover, the proposed multivariate PCI’s effectiveness is justified
through the implementation of multivariate data from different distributions with known
multivariate PCIs. It is worth noting that the statistical correlation between the multivariate
data affects the estimation of multivariate PCIs. Therefore, this paper applied an algorithm
that generates multivariate data, with a determined correlation. Consequently, applying the
proposed multivariate PCI algorithm on the generated data allows for examining how good
the proposed is. In addition to generating samples from different distributions, different
sample sizes from each sample were generated. In conclusion, the proposed multivariate
PCI algorithm was applied for different samples from different distributions and different
sample sizes. The results show the robustness of the proposed algorithm, which revealed
multivariate PCIs close to the actual PCIs of the used samples for different distributions and
different sample sizes. Finally, the research results indicate that the proposed multivariate
PCI outperforms the previous published algorithm in most cases.

Author Contributions: Conceptualization, M.A. and A.M.A.-A.; methodology, M.A. and A.Y.A.; soft-
ware, M.A.; validation, A.M.A.-A. and A.Y.A.; formal analysis, M.A.; investigation, M.A.; resources,
M.A.; data curation, M.A.; writing—original draft preparation, M.A.; writing—review and editing,
A.Y.A.; visualization, M.A.; supervision, A.M.A.-A. and A.Y.A.; project administration, A.M.A.-A.
and A.Y.A.; funding acquisition, A.M.A.-A. and A.Y.A. All authors have read and agreed to the
published version of the manuscript.
Funding: This study received funding from the Raytheon Chair for Systems Engineering. The authors
are grateful to the Raytheon Chair for Systems Engineering for funding.
Data Availability Statement: The data used in this study are available in this article.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Al-Refaie, A.; Bata, N. Evaluating measurement and process capabilities by GR&R with four quality measures. Measurement 2010,
43, 842–851.
2. Wang, F.-K.; Du, T. Using principal component analysis in process performance for multivariate data. Omega 2000, 28, 185–194.
[CrossRef]
3. Hahn, G.J.; Hill, W.J.; Hoerl, R.W.; Zinkgraf, S.A. The impact of Six Sigma improvement—A glimpse into the future of statistics.
Am. Stat. 1999, 53, 208–215.
4. Wang, F. Quality evaluation of a manufactured product with multiple characteristics. Qual. Reliab. Eng. Int. 2006, 22, 225–236.
[CrossRef]
5. Taam, W.; Subbaiah, P.; Liddy, J.W. A note on multivariate capability indices. J. Appl. Stat. 1993, 20, 339–351. [CrossRef]
6. Chen, H. A multivariate process capability index over a rectangular solid tolerance zone. Stat. Sin. 1994, 4, 749–758.
7. Shahriari, H.; Hubele, N.; Lawrence, F. A multivariate process capability vector. In Proceedings of the 4th Industrial Engineering
Research Conference, Institute of Industrial Engineers, Nashville, TN, USA, 24–25 May 1995.
8. Braun, L.J. New Methods in Multivariate Statistical Process Control(MSPC). Diskuss. Des Fachgeb. Unternehm. 2001, 1–12.
9. Castagliola, P.; Castellanos, J.-V.G. Capability indices dedicated to the two quality characteristics case. Qual. Technol. Quant.
Manag. 2005, 2, 201–220. [CrossRef]
10. Bothe, D.R. A capability index for multiple process streams. Qual. Eng. 1999, 11, 613–618. [CrossRef]
11. Boyles, R.A. Process Capability With Asymmetric Tolerances. Commun. Stat.-Simul. Comput. 1994, 23, 615–643. [CrossRef]
12. Davis, R.D.; Kaminsky, F.C.; Saboo, S.J. Process capability analysis for processes with either a circular or a spherical tolerance
zone. Qual. Eng. 1992, 5, 41–54. [CrossRef]
13. Yeh, A.B.; Bhattacharya, S. A robust process capability index. Commun. Stat.-Simul. Comput. 1998, 27, 565–589. [CrossRef]
Appl. Sci. 2023, 13, 11616 13 of 14

14. Veevers, A. Viability and capability indexes for multiresponse processes. J. Appl. Stat. 1998, 25, 545–558. [CrossRef]
15. Dianda, D.F.; Quaglino, M.B.; Pagura, J.A. Impact of measurement errors on the performance and distributional properties of the
multivariate capability index. AStA-Adv. Stat. Anal. 2018, 102, 117–143. [CrossRef]
16. Peruchi, R.S.; Rotela Junior, P.; Brito, T.G.; Largo, J.J.J.; Balestrassi, P.P. Multivariate process capability analysis applied to AISI
52100 hardened steel turning. Int. J. Adv. Manuf. Technol. 2018, 95, 3513–3522. [CrossRef]
17. Chatterjee, M.; Chakraborty, A.K. Unification of some multivariate process capability indices for asymmetric specification region.
Stat. Neerl. 2017, 71, 286–306. [CrossRef]
18. Dianda, D.F.; Quaglino, M.B.; Pagura, J.A. Distributional Properties of Multivariate Process Capability Indices under Normal and
Non-normal Distributions. Qual. Reliab. Eng. Int. 2017, 33, 275–295. [CrossRef]
19. Vasquez, M.; Ramirez, G.; Garcia, T. A multivariate process capability index based on non-conforming probability, an illustration
about monitoring the quality of a clarified water loop. Ing. UC 2016, 23, 319–326.
20. Pan, J.N.; Li, C.I.; Shih, W.C. New multivariate process capability indices for measuring the performance of multivariate processes
subject to non-normal distributions. Int. J. Qual. Reliab. Manag. 2016, 33, 42–61. [CrossRef]
21. Ciupke, K. Multivariate Process Capability Index Based on Data Depth Concept. Qual. Reliab. Eng. Int. 2016, 32, 2443–2453.
[CrossRef]
22. Ciupke, K. Multivariate Process Capability Vector Based on One-Sided Model. Qual. Reliab. Eng. Int. 2015, 31, 313–327. [CrossRef]
23. Mondal, S.C. A study of multivariate process capability indices in manufacturing processes. In Proceedings of the 2015 IEEE
International Conference on Industrial Engineering and Engineering Management (IEEM), Singapore, 6–9 December 2015; pp.
1382–1386.
24. Pan, J.-N.; Huang, W.K.C. Developing New Multivariate Process Capability Indices for Autocorrelated Data. Qual. Reliab. Eng.
Int. 2015, 31, 431–444. [CrossRef]
25. Siman, M. Multivariate Process Capability Indices: A Directional Approach. Commun. Stat.-Theory Methods 2014, 43, 1949–1955.
[CrossRef]
26. Zhang, M.; Wang, G.A.; He, S.G.; He, Z. Modified Multivariate Process Capability Index Using Principal Component Analysis.
Chin. J. Mech. Eng. 2014, 27, 249–259. [CrossRef]
27. Tano, I.; Vannman, K. A Multivariate Process Capability Index Based on the First Principal Component Only. Qual. Reliab. Eng.
Int. 2013, 29, 987–1003. [CrossRef]
28. Das, N.; Dwivedi, P.S. Multivariate Process Capability Index: A Review and Some Results. Econ. Qual. Control 2013, 28, 151–166.
[CrossRef]
29. Bashiri, M.J.M.; Amiri, A. A New Multivariate Process Capability Index under Both Unilateral and Bilateral Quality Characteristics.
Qual. Reliab. Eng. Int. 2012, 28, 925–941. [CrossRef]
30. Niavarani, M.R.; Noorossana, R.; Abbasi, B. Three New Multivariate Process Capability Indices. Commun. Stat.-Theory Methods
2012, 41, 341–356. [CrossRef]
31. Scagliarini, M. Multivariate process capability using principal component analysis in the presence of measurement errors.
AStA-Adv. Stat. Anal. 2011, 95, 113–128. [CrossRef]
32. Pan, J.N.; Lee, C.Y. New capability indices for evaluating the performance of multivariate manufacturing processes. Qual. Reliab.
Eng. Int. 2010, 26, 3–15. [CrossRef]
33. Ahmad, S.; Abdollahian, M.; Zeephongsekul, P.; Abbasi, B. Multivariate nonnormal process capability analysis. Int. J. Adv. Manuf.
Technol. 2009, 44, 757–765. [CrossRef]
34. Wen, D.C.; Lv, H. Multivariate Process Capability Index Based on the Additivity of Normal Distribution. In Proceedings of
the 2008 4th International Conference on Wireless Communications, Networking and Mobile Computing, Dalian, China, 12–14
October 2008.
35. Pearn, W.L.; Wang, F.K.; Yen, C.H. Multivariate capability indices: Distributional and inferential properties. J. Appl. Stat. 2007, 34,
941–962. [CrossRef]
36. Wang, C.H. Constructing multivariate process capability indices for short-run production. Int. J. Adv. Manuf. Technol. 2005, 26,
1306–1311. [CrossRef]
37. Wang, F.; Chen, J.C. Capability index using principal components analysis. Qual. Eng. 1998, 11, 21–27. [CrossRef]
38. Castagliola, P. Evaluation of non-normal process capability indices using Burr’s distributions. Qual. Eng. 1996, 8, 587–593.
[CrossRef]
39. Pearn, W.L.; Shiau, J.J.H.; Tai, Y.T.; Li, M.Y. Capability Assessment for Processes with Multiple Characteristics: A Generalization
of the Popular Index C-pk. Qual. Reliab. Eng. Int. 2011, 27, 1119–1129. [CrossRef]
40. Shiau, J.J.H.; Yen, C.L.; Pearn, W.; Lee, W.T. Yield-related process capability indices for processes of multiple quality characteristics.
Qual. Reliab. Eng. Int. 2013, 29, 487–507. [CrossRef]
41. Abbasi, B.; Akhavan Niaki, S.T. Estimating process capability indices of multivariate nonnormal processes. Int. J. Adv. Manuf.
Technol. 2010, 50, 823–830. [CrossRef]
42. Clements, J.A. Process Capability Calculations For Non-Normal Distributions. Qual. Prog. 1989, 22, 95–97.
43. Somerville, S.E.; Montgomery, D.C.J. Process capability indices and non-normal distributions. Qual. Eng. 1996, 9, 305–316.
[CrossRef]
44. Zimmer, L. Process Capability Indices in Theory and Practice. Technometrics 2000, 42, 206–207. [CrossRef]
Appl. Sci. 2023, 13, 11616 14 of 14

45. Deleryd, M. On the gap between theory and practice of process capability studies. Int. J. Qual. Reliab. Manag. 1998, 15, 178–191.
[CrossRef]
46. Wu, H.; Wang, J.; Liu, T. Discussions of the Clements-based process capability indices. In Proceedings of the 1998 CIIE National
Conference; pp. 561–566. Available online: https://www.ciie.org/zbh/en/activities/thirdciie/agenda/ (accessed on 8 September
2023).
47. Kotz, S.; Johnson, N.L. Process capability indices—A review, 1992–2000. J. Qual. Technol. 2002, 34, 2–19. [CrossRef]
48. Liu, P.-H.; Chen, F.-L. Process capability analysis of non-normal process data using the Burr XII distribution. Int. J. Adv. Manuf.
Technol. 2006, 27, 975–984. [CrossRef]
49. Piao, C.; Zhi-Sheng, Y. A systematic look at the gamma process capability indices. Eur. J. Oper. Res. 2018, 265, 589–597.
50. Li, C.-J.; Deng, W.-P.; Cao, Y.-Y.; Bao, Y. Process capability analysis in non-normality based on Box-Cox transformation and
Johnson transformation. J. Qiqihar Univ. Nat. Sci. Ed. 2015. Available online: https://www.scinapse.io/papers/2383696946
(accessed on 8 September 2023).
51. Bernardo, J.M.; Irony, T.Z.J. A general multivariate Bayesian process capability index. J. R. Stat. Society. Ser. D 1996, 45, 487–502.
[CrossRef]
52. Wang, F.-K.; Hubele, N.F.J. Quality evaluation using geometric distance approach. Int. J. Reliab. Qual. Saf. Eng. 1999, 6, 139–153.
[CrossRef]
53. Pal, S.J. Evaluation of nonnormal process capability indices using generalized lambda distribution. Qual. Eng. 2004, 17, 77–85.
[CrossRef]
54. Chen, K.-S.; Hsu, C.-H.; Wu, C.-C. Process capability analysis for a multi-process product. Int. J. Adv. Manuf. Technol. 2006, 27,
1235–1241. [CrossRef]
55. Niaki, S.T.A.; Abbasi, B.J. Skewness reduction approach in multi-attribute process monitoring. Commun. Stat.-Theory Methods
2007, 36, 2313–2325. [CrossRef]
56. Perakis, M.; Xekalaki, E. On the Implementation of the Principal Component Analysis–Based Approach in Measuring Process
Capability. Qual. Reliab. Eng. Int. 2012, 28, 467–480. [CrossRef]
57. Dharmasena, L.; Zeephongsekul, P. A new process capability index for multiple quality characteristics based on principal
components. Int. J. Prod. Res. 2016, 54, 4617–4633. [CrossRef]
58. Wang, S.; Wang, M.; Fan, X.; Zhang, S.; Han, R. A multivariate process capability index with a spatial coefficient. J. Semicond.
2013, 34, 026001. [CrossRef]
59. Gu, K.; Jia, X.; Liu, H.; You, H. Yield-based capability index for evaluating the performance of multivariate manufacturing process.
Qual. Reliab. Eng. Int. 2015, 31, 419–430. [CrossRef]
60. Tiwari, V.; Singh, N. Process capability index for bivariate exponentially distributed quality characteristics and its sampling
properties. Commun. Stat.-Theory Methods 2017, 46, 11099–11109. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.