You are on page 1of 16

Quality Engineering

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/lqen20

Application of a proposed reliability analysis


multivariate capability index on manufacturing
processes

Rister Barreto & Herrera Acosta Roberto

To cite this article: Rister Barreto & Herrera Acosta Roberto (2021): Application of a proposed
reliability analysis multivariate capability index on manufacturing processes, Quality Engineering,
DOI: 10.1080/08982112.2021.1973035

To link to this article: https://doi.org/10.1080/08982112.2021.1973035

Published online: 10 Sep 2021.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=lqen20
QUALITY ENGINEERING
https://doi.org/10.1080/08982112.2021.1973035

Application of a proposed reliability analysis multivariate capability index on


manufacturing processes
Rister Barreto and Herrera Acosta Roberto
Engineering Faculty, Universidad del Atlantico, Barranquilla, Colombia

ABSTRACT KEYWORDS
n the manufacturing process of any product intervene several factors associated with the Process capability index;
monitored quality characteristics. These factors must be examined with adequate Statistical multivariate analysis;
Quality Control tools to provide estimates of product specification compliance. For this pur- principal component
analysis; product reliability;
pose, Process Capability Indices are commonly applied to manufacturing processes, assum- gamma distribution
ing that the quality variables are independent, and the data is normally distributed.
Additionally, an increasingly prominent quality characteristic across manufacturing industries
is product reliability. This paper explores the advantages of using multivariate capability
indices that include product reliability as a key quality criterion.

Introduction (Wooluru, Swamy, and Nagesh 2014). Generally, a


process needs to be under statistical control before
Process Capability Analysis is an engineering tech-
evaluating its capability, which means that if there are
nique that facilitates the evaluation of product compli-
ance with technical specifications and tolerance assignable causes for the out-of-control signs in the
intervals. By monitoring the variables historically or process, they must be removed in order to guarantee
lineally, process capability analysis provides an esti- that only chance causes are responsible for variations
mate of the product quality conditions and simplifies in the system (Shinde and Katikar 2012). Otherwise,
the evaluation of product compliance with quality the results from the capability analysis may be incor-
standards. The importance of meeting these standards rect and they can lead to inadequate estimations about
is that they often reflect customer requirements and process performance.
lead to overall high-quality production. Consequently, Through Process Capability Indices (PCIs), it is
in order to achieve final product quality, it is neces- possible to summarize process capability information
sary to ensure that every production stage adheres to into numerical data that compares the behavior of a
the specifications. Through Statistical Quality Control product or process characteristic to its engineering
(SQC) methods, various statistical principles can be requirements (Steiner, Bovas, and MacKay 1998).
applied at the stages of design and manufacturing of a Additionally, a process capability index often serves
product, as well as services. Thus, SQC techniques the purpose of communicating the specifications set
have made great contributions to companies that by the customers, thus guaranteeing that their needs
require improvement in their mass production proc- are ultimately met. In general, the indices are used to
esses (Shinde and Katikar 2012). report on the performance of the process and its cap-
Furthermore, the administration and implementa- ability to produce parts that will meet the technical
tion of process capability studies can greatly influence specifications. Despite these advantages, implementing
the quality of engineering since the analysis results PCIs requires that certain process conditions are met,
can be used for new design applications, inspection which sometimes constitutes a limitation. In order to
planning and evaluation techniques. This tool can also correctly utilize this tool in a manufacturing process,
prevent future defects in the production process the data must follow a normal distribution and the
through design improvement, knowledge of machine observations must be independent. These restrictions
or process limitations and process factor control are especially disadvantageous in processes where the

CONTACT Rister Barreto rjbarreto@mail.uniatlantico.edu.co Engineering Faculty, Universidad del Atlantico, Kra 35 #47-105, Barranquilla, Atlantico
80005, Colombia.
ß 2021 Taylor & Francis Group, LLC
2 R. BARRETO AND R. HERRERA

influencing variables are closely related, which is the 5. The addition of key quality criteria, such as reli-
case for most real-world production systems. In these ability, to the univariate indicator, Herrera et
instances, it is necessary to resort to Multivariate al. (2018).
Process Capability Indices (MPCIs) that take these 6. Other authors have presented different proposals
correlations into account. In SQC literature, several for multivariate capability indices, such as (Shariai
multivariate capability index proposals have been pre- and Abdollahzadeh 2009).
sented in order to provide a process performance
measure aligned with the design specifications of each Principal components multivariate capability index
variable involved in the process.
In this study, a Multivariate Capability Index is In the case of bilateral specifications, Wang and Du
proposed, based on the analysis of product compli- (2000) presented the formulas for the r initial compo-
ance with the technical specifications and with the nents as follows
" #1
addition of a new factor that takes into consideration Y
r   r
the reliability of the product based on product war- MCPU ¼  Cpu Yi 
[1]
i¼1
ranty promises and expected shelf-life. This is com-
pared to a traditional MPCI, based on the Principal " #1r
Y
r  
Component Analysis (PCA) as introduced by Chen, MCPL ¼ Cpl Yi  [2]
Pearn, and Lin (2003). The rest of the paper is com- i¼1

posed by a Theoretical Framework Section, which Where Cpu Yi and Cpl Yi are the process capability
focuses on introducing the concepts of Multivariate index values for each of the ith principal components.
Process Capability Indices, the PCA technique, prod- Let PR and SR be the Process Region and Specification
uct reliability background and the Gamma Region, respectively. For processes where the PR is cen-
Distribution. Next, an analysis on normality and auto- tered on the SR, the bilateral process capability indices
correlation assumptions is presented to explain the are defined as Cpu ¼ ðUSL  lÞ=3r and Cpl ¼
role these factors play in both the traditional and the ðl  LSLÞ=3r, where l is the mean value of the prod-
proposed model. Two different case studies are uct characteristic sample. However, this approach does
included to highlight the contribution of this study to not consider other scenarios where the target value is
process capability analysis and lastly, the conclusions not centered on the SR. The Cpk index considers the
and future work recommendations explain the main minimal distance between the midpoint of the PR and
findings of this research and provide guidance for the its closer specification limit. The measure is defined as
application of the proposed model on cases that do Cpk ¼ fCpl , Cpu g [3]
not adhere to the standard conditions.
The Principal Component Analysis is used to
obtain a new set of variables (Principal Components)
based on the individual characteristics of a product
Theoretical framework
(Jackson 1980). The multivariate capability index
Multivariate capability index using the PCA method was introduced by Wang and
Chen (1998) and it is a measure that evaluates process
According to Shinde and Khadse (2009), and Herrera
capability considering the relationship between the
et al. (2018), there are generally six ways to define
quality characteristics studied. Let Cp;Yi represent the
multivariate capability indices:
univariate measure of process capability for the ith
principal component, then
1. The ratio of the tolerance and process regions as
" #1
proposed by Taam, Subbaiah, and Liddy (1993). Yr   r
MCp ¼ Cp;Y  [4]
2. The evaluation of the non-conforming products i
i¼1
proportions as presented by Taam, Subbaiah, and
Liddy (1993), Castagliola et al. (2009), Bothe Note that Cp;Yi can be replaced with Cpk;Yi , Cpm;Yi
(1991) and Wierda (1994). and Cpmk;Yi , obtaining similar indices for the trad-
3. The application of Principal Component Analysis, itional univariate capability measures. K.S. Chen,
as suggested by Wang and Chen (1998). Pearn, and Lin (2003) presented process capability
4. The application of Functional Data Analysis indices to evaluate v characteristics, assuming nor-
(FDA) (Guevara and Vargas 2016). mally distributed data and independent variables. Let
QUALITY ENGINEERING 3

Figure 1. Yield autocorrelation test example.

MCpk be the capability index for multiple variables i ¼ 2 adds the product reliability model criterion. A,
based on the characteristics of the process, its calcula- on the other hand, is a constant defined as:
tion is given by " v #
Y   
8" #9 A¼ 2U 3Cpkj  1 þ 1 [8]
< Q
> v     >
2U 3Cpkj  1 þ 1 = j¼1
j¼1
MCpk ¼ U1 > >
1
: ; [5] The values of Cpkj are calculated using Eqs. [3] and
3 2 [6] for each quality characteristic. A very important
Where v is the total number of variables studied factor to consider for the constant A is the presence
and Cpkj is the value of the jth characteristic for of autocorrelation. Correlation between variables is
j ¼ 1, 2, :::N: one of the favorable aspects of using MPCIs since the
  simultaneous analysis of multiple quality characteris-
Cpkj ¼ min Cpuj ; Cplj [6]
tics offer a global overview of the capability of the
Equation [6] is used as a locating process capability process. Autocorrelation, on the other hand, can
index, based on the indices for upper and lower speci- potentially influence the results of the capability ana-
fication limits. lysis depending on the level of autocorrelation and the
specific statistical measures that are being imple-
mented in the process. A study presented by Guevara
Reliability analysis multivariate capability index
and Vargas (2007) analyzed the effects of autocorrel-
Assuming normality in the data distribution and ation on traditional capability indices Cp , Cpk , Cpm
mutual independence in the observations for each and Cpmk and concluded that the higher the autocor-
quality characteristic, the proposed multivariate cap- relation in the observations, the lower the capability
ability index is a modification of the index presented index. Additionally, they found that autocorrelation
by Chan et al. (2003). The new index includes reliabil- does not affect the expected value of sample mean but
ity RðtÞ as an important factor in process performance does impact the expected value of the standard error,
evaluation, given that reliability is an important criter- which increases as sample size increases.
ion that is overlooked in current multivariate capabil- Despite the fact that autocorrelation is prevalent in
ity index proposals. Thus, the proposed index is many continuous manufacturing processes (e.g., in the
calculated as follows chemical and pharmaceutical industry), the analysis of
 
this factor is often disregarded to prioritize normality
1 1 A 1
MCpCR ¼ w1 U þ w2 U ½RðtÞ [7] distribution tests and in-control process assessment
3 2
(Sun, Shengxian, and Zhihui 2010). As mentioned
Where wi is the assigned weight for product com- before, the proposed model sustains the normality and
pliance with the specifications, as well as with product mutual independence premises; therefore, normality
P
reliability, where 2i¼1 wi ¼ 1: In this case, i ¼ 1 uses and autocorrelation tests were applied to the collected
the quality criterion based on the specifications and data. The results generally show that the hypothesis of
4 R. BARRETO AND R. HERRERA

normality and statistically insignificant autocorrelation final decision on which method to apply will depend
factors could not be rejected with P-values over 0.05 on the conditions of the process. For highly correlated
and 95% confidence. Figure 1 shows an autocorrel- product characteristics, a multivariate approach will
ation test example for one of the variables studied, be more effective in assessing process capability than
where it can be observed that the autocorrelation fac- individual indices for each characteristic.
tor does not surpass the significance level limit with a De-Felipe and Benedito (2017) describe a method
lag k ¼ 1: Similarly, the other variables presented to determine which PCI should be used for any given
results that allow the autocorrelation hypothesis to be process. Depending on the characteristics of the data
rejected. However, the recommendations for cases available and the expected output of the process cap-
outside these parameters will be discussed in a later ability analysis, decision makers should decide if the
section of this paper. process will be treated as a multivariate or univariate
process. For multivariate settings, there are different
Multivariate capability indices analysis approaches to consider based on four different crite-
and comparison ria: calculation complexity, globality, relation to NCP
and robustness. The PCA technique offers the advan-
Process Capability Analysis often entails assessing tage of robustness and globality, which is the ability
processes and products based on key quality charac- to synthesize the capability of multiple characteristics
teristics, usually guided by technical specification
in a single index, thus providing insights into the
standards. Since a product rarely depends on a single
overall quality of the process.
characteristic that defines its quality, most real-life sit-
The criteria to determine whether the process cap-
uations involve evaluating several quality variables.
ability index chosen indicates a positive or negative
Traditionally, companies have resorted to obtain uni-
result can be established by company policy, depend-
variate PCIs for each product characteristic they wish
ing on how stringent the measure should be for any
to assess. This practice, however, can lead to inaccura-
given process. However, a widely accepted perform-
cies in the capability analysis because the influence
ance measure is
each variable has on the others is not considered (de-
Felipe and Benedito 2017). MCp > 1:0 the process capability is optimal
Multivariate capability indices and, more specific-
ally, the Principal Component Analysis methodology MCp ¼ 1:0 process capability is acceptable
solve this problem by evaluating all the product char-
acteristics simultaneously and accounting for all pro- MCp < 1:0 process capability is inadequate
cess variability globally. The PCA technique
Other authors suggest an interval-based approach
transforms the v characteristics of a product into a set
for more specificity, e.g., Tsai and Chen (2006) sug-
of linear functions of the original measurements,
gest the following to assess process capability for uni-
called principal components. Each component is a
variate PCIs:
combination of the original variables and account for
the variation in the characteristics studied (Wang and Cp > 2:00 Super Excellent
Chen 1998). Typically, the first principal component
contains the maximum variation of each product 1:67  Cp  2:00 Excellent
characteristic in the process. In this sense, the PCA
technique evaluates the capability of a multivariate 1:33  Cp  1:67 Satisfactory
process by relating the variability of multiple product
1:00  Cp  1:33 Capable
characteristics to their acceptable standard measures.
It is important to note that the PCA methodology 0:67  Cp  1:00 Inadequate
is a combination of univariate PCIs that are weighted
in each component based on the contribution each Cp < 0:67 Poor
variable makes to the overall variation in the process.
Therefore, these methodologies are complementary It is the responsibility of decision-makers to estab-
rather than mutually exclusive. Consequently, a posi- lish the criteria for process capability index evaluation.
tive multivariate PCI measure is generally the result of In this case, both the traditional and proposed indices
acceptable capability indices for each product charac- (MCp and MCpR ) will be considered under Tsai and
teristic and vice versa. It is important to note that the Chen’s scale, applied to the multivariate setting.
QUALITY ENGINEERING 5

Product reliability determines the shape of the distribution or the rate in


the case of queuing models. The standard gamma
Product reliability refers to the performance of a product
probability density function is
over time. A more technical definition, as explained by
Blischke and Murthy (2000), is that the reliability of a xa1 ex
f ðx; aÞ ¼
product is the probability that, under normal conditions, CðaÞ
the product will perform its expected functions for a spe-
A random lifetime variable is distributed according
cific period of time. This definition illustrates the import-
to a Gamma with parameters a and k, Gða, kÞ, when
ance of product reliability for both the customers and
its density function is given by
the manufacturers, as it presents new challenges for pro-
duction systems that must guarantee reliability as part of ka t a1 kt
f ðt Þ ¼ e t  0; a, k > 0
the quality dimensions of the final products. There are CðaÞ
different factors that influence final product reliability,
Where a is the shape parameter, k is the scale par-
which is normally embedded in the manufacturing pro-
ameter and C is the Gamma function, defined as
cess and must be analyzed at every stage of production. ð1
Consequently, reliability can be influenced by design reli-
CðaÞ ¼ xa1 ex dx 8 a > 0
ability, due to assembly errors and component noncon- 0
formance; sales reliability, which can be affected by
When a ¼ 1, then the gamma distribution becomes
storage time and handling or reliability performance,
the exponential distribution with k ¼ b1 : This prob-
once the product is sold and the unit is either stored or
ability density function will be utilized in the pro-
put into operation immediately (Murthy 2007).
posed model to analyze product time-based quality
In the current market, constantly expanding due to
characteristics to compare the results to the index
technology advancement, reliability has become a
obtained applying the traditional PCA methodology.
quality characteristic expected by customers. It is the
responsibility of manufacturers to offer warranties for
Case Study I: Levocetirizine
their products in order to reassure customers that
they will perform well over time. This increasingly
Levocetirizine is an antihistamine, used for treating
stringent market expectation involves expenditure on
the symptoms of allergic diseases such as Seasonal
research and development for the companies (Murthy
Allergic Rhinitis (including ocular symptoms) and
and Jiang 2009), who are also responsible for failures
Chronic Idiopathic Rhinitis in patients from six
within the warranty period and their rectification.
months old on. The development of analytic tests for
Based on these considerations, product reliability con-
determining pH levels, density, active ingredient
stitutes not only an advantage in the current market
assessment and sucrose content in product manufac-
but also an obligation for manufacturers. Using reli-
ability as a factor in process capability analysis and turing aim to determine the compliance level with the
indices can result in an overall more stringent criteria specifications set for each of these variables, according
for high-quality performance and therefore, product to preestablished standards. An evaluative research
quality improvement and higher customer satisfaction. was conducted, where data for each of the studied
variables were collected to evaluate the performance
of the Levocetirizine manufacturing process. The data
Gamma distribution sample is summarized in Table 1.
In probability theory, the gamma distribution is a two- Table 2 presents the Upper and Lower Specification
parameter group of continuous probability distributions, Limits for each one of the variables studied.
which can be used to analyze the lifetime of components
and a variety of queuing models (Varshini et al. 2018). In Methodology application
this case, the reliability parameter included in the multi-
variate capability index proposal (see Eq. [7]) follows a Traditional method: Principal component analysis
gamma distribution, for which the probability density The Principal Component Analysis (PCA) technique,
function is defined as as proposed by Wang and Chen (1998), was applied
1 to the collected data in order to calculate the multi-
xa1 eb
x
f ðx; a, bÞ ¼ a
b CðaÞ variate capability index that constitutes the basis of
Where a and b are parameters with positive real the proposed method. The main purpose of the PCA
values. The parameter a is a scale factor, while b is to obtain a reduced number of linear combinations
6 R. BARRETO AND R. HERRERA

Table 1. Data sample for quality characteristics in levocetiri- Table 3. Eigenvalues and variance percentage per component.
zine manufacturing process. Variance Cumulated
pH Density Active ingredient assessment Sucrose content Component Eigenvalue percentage (%) percentage (%)
2.51 0.998 99.534 62.072 1 2.374 59.347 59.347
2.58 0.9941 99.523 63.939 2 1.102 27.545 86.891
2.63 1.0816 99.512 65.045 3 0.511 12.764 99.656
2.72 1.0339 99.492 67.261 4 0.014 0.344 100.00
2.74 1.0403 99.462 68.022
2.74 1.0615 99.616 68.075 through the PCA technique. The total variance per-
2.82 1.2663 99.492 68.103
2.76 1.0555 99.571 67.965 centage for each component is obtained by calculating
2.75 1.0385 99.417 67.827 the total weight of each component compared to the
2.75 1.0801 99.502 68.021
2.78 1.0152 99.406 68.212
total sum of the eigenvalues. If the second component
2.78 0.9972 99.416 68.394 is also included, they represent a cumulated 86.9% of
2.79 1.0913 99.465 68.883 the total variance.
2.79 1.0497 99.451 67.972
2.79 1.0668 99.426 68.049 The values in Table 4 are the coefficients for the
principal component equations, e. g., the first princi-
Table 2. Levocetirizine variable specifications. pal component is represented by the equation:
Upper and lower Unit of 0:639189  pH þ 0:333545  Density  0:321686
Variable specification limits measurement
pH 2.50  3.50  Assessment þ 0:63762  SucroseContent:
Density 1.00  1.26 g/ml
Active ingredient 90  110 The coefficients for each component equation have
assessment
Sucrose content 65  70 % interesting connotations, it can be observed that the first
Reliability 2 years component is weighted in a positive direction for the
variables pH, Density and Sucrose Content and in a
of the four studied variables that explain the variabil- negative direction for the Active Ingredient Assessment.
ity of the data. Thus, the starting point of this tech- The second and third component, on the contrary, show
nique consists of the mean vector (l) and variance- a strong positive direction for the Assessment value.
covariance matrix (S) calculations. The number of components to extract for the ana-
2 3 lysis depends mostly on the investigation criteria set by
2:72867
6 1:05800 7 the researchers. Different rules may be applied to make
l¼6 7
4 99:4857 5 the decision, such as extracting the components with
eigenvalues larger than one or plotting the sedimenta-
67:1893
tion of the components as shown in Figure 2. In this
And case, the first two components are higher than one [1]
  and explain 87% of the variance. Thus, the analysis will
 0:00096429 0:00100475 0:00068679 0:01511320 

 0:00078468 0:00373511  proceed using these factors. The normalized eigenvec-
 0:00100475 0:00408036
S¼ 
 0:00068679 0:00078468 0:00322693 0:00178746  tors using the values in Table 4 are presented below:
 
 0:00178746 0:40795000 
0:01511320 0:00373511 ^ 1 ¼ ð0:639189; 0:333545;  0:321686; 0:613762Þ
u
Table 3 summarizes the results obtained from the
PCA calculations of eigenvalues, applying the eigen- ^ 2 ¼ ð0:014799; 0:700623; 0:713025; 0:022448Þ
u
decomposition theorem. Figure 3 is a visual representation of the position
Consequently, the eigenvectors associated with the of each variable on the cartesian plane. The variables
principal components are described in the follow- in the further positions from the reference points (0),
ing matrix: contribute more to the component. Once the
extracted principal components and its corresponding
0.00096429 0.00100475 –0.00068679 0.01511320
vectors have been determined, the next step in the
0.00100475 0.00408036 –0.00078468 0.00373511 PCA technique is to calculate the linear combinations
–0.00068679 –0.00078468 0.00322693 –0.00178746 of the mean the specifications vectors, using the coef-
0.01511320 0.00373511 –0.00178746 0.40795000
ficients of the normalized vectors. Thus, the results
obtained for each component are determined by
According to the eigenvalues obtained, the first Xn

component contains 59.3% of the total variance in the LSL : LYi ¼ LSLvj  Dij
i;j
four quality characteristics that are being evaluated
QUALITY ENGINEERING 7

Table 4. Component weight.


Variable Component 1 Component 2 Component 3 Component 4
pH 0.639189 0.014799 0.209187 0.739905
Density 0.333545 0.700623 –0.617753 –0.127503
A.I. assessment –0.321686 0.713025 0.616550 0.089325
Sucrose content 0.613762 –0.022448 0.441008 –0.654450

Figure 2. Component sedimentation in PCA for Case I.

Table 5. Process capability indices for each variable.


Variable Cpu Cpl Min fCpu , Cpl g
pH 2.9441 0.8728 0.8728
Density 0.9760 0.2948 0.2948
A. I. assessment 58.2380 52.5403 52.5403
Sucrose content 0.4857 0.3784 0.3784

Figure 3. Components weight biplot Case I.

X
n Similarly, the centers of specifications for each com-
USL : UYi ¼ USLvj  Dij ponent are calculated with the formula
i;j
X n
Center : PCmYi ¼ Cvj  Dij
Where LSLvi and USLvi are the lower and upper i;j
specifications limits for each quality characteristic vi : Where Cvj are the centers of specifications for each
On the other hand, Dij are the component weight quality characteristic. Thus, the first component linear
coefficients for each component i and characteristic j: calculations are
8 R. BARRETO AND R. HERRERA

Figure 4. Component sedimentation in PCA for Case II.

LSL : LY1 ¼ ð2:5  0:639189 þ 1:0  0:333545 þ 90 variation to consider a non-centered PR. Note that r ¼ 2,
 ð0:321686Þ þ 65  0:613762Þ defined by the two components extracted from the PCA
results.
¼ 12:8743
1
MCpk ¼ ½0:2858  2:27672 ¼ 0:8063
USL : UY1 ¼ ð3:5  0:639189 þ 1:25  0:333545 þ 110
The Cpk;Yi values were calculated with the equation
 ð0:321686Þ þ 70  0:613762Þ

UYi  CYi CYi  LYi


¼ 10:2319 Cpk, Yi ¼ min ,
3r 3r

The specification limits for the second component Y2 are Where r is the square root of each component
calculated using the corresponding coefficients for this eigenvalue and CYi are the central values of each com-
component and the specification criteria presented in ponent, obtained by evaluating the specifications cen-
Table 3. tral values in the normalized vector function.
  

LSL : LY2 ¼ ð2:5  0:014799 þ 1:0  0:700623 þ 90  10:2319  11:5531   10:2319  11:5531 
Cp, Yi ¼ min  pffiffiffiffiffiffiffiffiffiffiffi   pffiffiffiffiffiffiffiffiffiffiffi 
3  2:374 ,  3  2:374 
 0:713025 þ 65  ð0:022448ÞÞ
¼ 0:2858
¼ 63:4507
  

USL : UY2 ¼ ð3:5  0:014799 þ 1:25  0:700623 þ 110  77:7889  70:6233Þ   70:6233  63:4507 
Cp, Yi ¼ min  pffiffiffiffiffiffiffiffiffiffiffi , 
 p ffiffiffiffiffiffiffiffiffiffiffi 

3  1:102 3  1:102
 0:713025 þ 70  ð0:022448ÞÞ
¼ 2:2775
¼ 77:7889
Thus far, the Levocetirizine manufacturing process
Then, the centers of specifications for each compo- capability has been evaluated using only the traditional
nent are calculated as multivariate capability analysis proposed by Chen. The
PCmY1 ¼ ð3:0  0:639189 þ 1:1  0:333545 þ 100 Principal Component Analysis establishes that a process
is capable of complying with product specifications if the
 ð0:321686Þ þ 67:5  0:613762Þ global value is greater than one [1] (MCpk > 1), while
¼ 11:5548 the contrary is usually a sign of underlying quality issues
in the process that will likely result in defective final
PCmY2 ¼ ð3:0  0:014799 þ 1:1  0:700623 þ 100 products. In this case, the global value obtained for the
 0:713025 þ 67:5  ð0:022448ÞÞ Levocetirizine manufacturing process is 0.8063, which
indicates that if the correlation between the variables
¼ 70:6223
involved in the process (pH, Density, Active Ingredient
Through the principal component analysis, the global and Sucrose Content) is considered, the process is unable
capability index is calculated using Eq. [4] with the Cpk;Yi to produce specification-complying items.
QUALITY ENGINEERING 9

Table 6. Reliability analysis multivariate capability index; Case I.


Form Scale Time (years)
Parameters 0.9244 5025 2
Reliability RðtÞ 0.9993
Weight Specifications compliance ðw1 Þ Product reliability ðw2 Þ
0.5 0.5
Reliability capability index MCpR 0.6317

Suggested approach: Product reliability analysis 1 


MCpR ¼ 0:5  U1 ½0:7298 þ 0:5  U1 ½0:9996575
3
In the previous section, the PCA methodology was
applied to the collected data presented in Table 1. The ¼ 0:6317
same dataset was analyzed using the proposed
approach i.e., the Reliability Analysis Multivariate The overall results are presented in Table 6, indicat-
Capability Index. First, the differentiating factor of the ing the different parameters used and their value for
proposed method was calculated. As mentioned each factor included in the process capability analysis.
before, product reliability is an important criterion The global reliability analysis capability index for
that directly influences customer satisfaction and con- the Levocetirizine manufacturing process is 0.6317.
stitutes an obligation for manufacturers that want to Similar to the traditional PCA-based multivariate cap-
ensure product performance over time. Using the ability index, the expected value for the proposed reli-
gamma distribution density function with the parame- ability index describing an optimized manufacturing
ters of shape and scale defined as a ¼ 0:92 and b ¼ process is MCpR > 1:0: However, many companies
5025, the probability of the product having a lifetime nowadays have established a MCpR > 1:2 mark to
less than or equal to two years (t ¼ 2) is consider the process performance acceptable. This
f ð2; 0:92, 5025Þ ¼ 0:0007412 indicates that the overall capability performance of
this process is not capable of producing specification-
Thus, the resulting reliability factor, considering complying items with customer-appealing product
that the warranty promises a lifetime greater than two reliability assurance.
years is
Rð2Þ ¼ 1  0:0007412 ¼ 0:9992588 Case Study II: Latex Paint
The proposed method also takes into account the Three main quality variables were studied in the
parameter of product compliance with the specifica-
manufacturing process of a water-based paint to ana-
tions A, for which it is necessary to calculate the pro-
lyze its capability: performance, thickness, and viscos-
cess capability indices Cpu and Cpl for each one of the
ity. Four different paint batches were sampled in
variables included in the study i.e., pH, Density,
order to collect the observations presented in Table 7.
Active Ingredient Assessment and Sucrose Content.
The Principal Component Analysis technique was
The results are summarized in Table 5.
  applied to the data in order to determine a traditional
Note that Cpkj ¼ min Cpuj ; Cplj was also applied to
the indices, using Eq. [6] to find the values for each multivariate capability index based on these
variable. Finally, through Eq. [8], the product compli- measurements.
ance factor obtained is Table 8 presents the Upper and Lower Specification
Limits set by the manufacturer based on previous
A
¼ 0:7298 studies regarding the performance of the paint and
2
the requirements of the customers.
In this case, the weight wi was 0.5 for both the
product compliance with the specifications and
product reliability, but this will depend on the criteria Methodology application
of the manufacturer and the importance they assign For the second case study, the same analysis method-
to each of these factors in the evaluation process. ology was applied to the collected data. First, the prin-
Once these parameters have been established, the cipal component analysis technique was implemented,
proposed multivariate capability index was calculated using the traditional method proposed by Chen in
using Eq. [7]: order to obtain a PCA-based Multivariate Capability
10 R. BARRETO AND R. HERRERA

Table 7. Collected data from water-based paint manufactur- Table 9. Eigenvalues and variance percentage per component.
ing process. Variance Cumulated
Yield (l=m2 Þ Viscosity ðPa  sÞ Thickness (mils)a Component Eigenvalue percentage (%) percentage (%)
11.84 25.20 12.94 1 1.8992 63.3050 63.3050
12.03 22.05 13.14 2 1.0000 33.3350 96.6410
12.79 19.88 13.34 3 0.1008 3.3590 100.000
12.83 21.06 13.91
12.83 22.01 13.91
13.00 24.05 14.09
13.18 23.11 14.28
13.00 24.04 14.22 Table 10. Component weight table.
13.15 21.41 14.26 Variable Component 1 Component 2
13.18 23.15 14.20 Yield 0.7071 –0.0007
13.26 25.95 14.37 Viscosity –0.0323 0.9989
12.52 21.46 13.58 Thickness 0.7063 0.0463
12.86 20.43 13.94
13.32 25.22 14.20
12.94 22.87 14.14
13.23 22.75 14.51 The eigenvalues calculated applying the eigen-decom-
13.25 24.67 14.26
13.26 24.33 14.37 position theorem are presented in Table 9.
13.14 23.62 14.24 These results show that the first component pos-
14.19 21.57 15.58
12.92 22.86 14.00
sesses the majority of the variance in the dataset with
13.25 23.77 14.36 63.30%. With over half of the total variance of the
12.79 21.45 13.86
13.30 22.13 13.42
three characteristics in the first component, it is
12.97 23.22 14.16 already noticeable that only the first two components
13.62 24.27 14.46 are relevant to the PCA study.
12.45 27.25 13.39
13.55 21.50 14.68 The values in Table 10 are the coefficients for the
12.56 22.81 13.55 principal component eigenvectors e. g., the first princi-
13.04 25.74 14.12
a
Measured for two layers of paint.
pal component is represented by the equation: 0:7071 
Yield  0:0323 Viscosity þ 0:7063  Thickness:
Table 8. Upper and lower specification limits for water-based As explained in the previous case study, the deci-
paint manufacturing. sion of how many principal components will be
Variable Upper and lower specification limits Unit of measurement defined, relies on the criteria of the investigation.
Yield 12.00  15.00 (l=m2 Þ Figure 4 shows that the first two components account
Viscosity 20.00  24.00 ðPa  sÞ
Thickness 12.00  15.00 ðmilsÞ
for over 96% of the variance in the sample and there-
fore, they will be enough to proceed with the PCA
methodology. The normalized eigenvectors are pre-
Index. Then, the proposed Commercial Multivariate sented below.
Capability Index was applied to the same dataset,
including the Reliability Factor based on a separate set ^ 1 ¼ ð0:7071;  0:0323; 0:7063Þ
u
of quality characteristic measurements of the same
paint manufacturing process. ^ 2 ¼ ð0:0007; 0:9989; 0:0463Þ
u
The biplot in Figure 5 shows that Thickness and
Traditional method: Principal component analysis Yield influence mostly component 2, while Viscosity
is the main characteristic driving the first component.
Similar to the first case study, the mean vector (l) Next, the linear combinations of each component
and variance-covariance matrix (S) were calculated for were calculated as follows:
the quality characteristics studied. The Upper and Lower Specifications Limits for the
2 3 first component Y1 are calculated as
13:01
l ¼ 4 23:13 5 LSL : LY1 ¼ ð12  0:0771 þ 20  ð0:0323Þ þ 12
14:05  0:7063Þ ¼ 8:7548

And, USL : UY1 ¼ ð15  0:0771 þ 24  ð0:0323Þ


  þ 15  0:7063Þ ¼ 10:9758
 0:2036 0:0326 0:2082 

S ¼  0:0326 3:0152 0:0006  The specification limits for the second and third com-
 0:2082 0:0006 0:2639  ponent Y2 are calculated using the corresponding
QUALITY ENGINEERING 11

Figure 5. Components weight biplot Case II.


  
coefficients for the component and the specification 24:6576  22:5683Þ 22:5683  20:5252
Cp, Yi ¼ min  pffiffiffiffiffiffiffiffiffiffiffiffiffi ,  pffiffiffiffiffiffiffiffiffiffiffiffiffi g ¼ 0:6810
criteria presented in Table 10. 3  1:0000 3  1:0000

LSL : LY2 ¼ ð12  ð0:0007Þ þ 20  0:9989 The global value 0.3532 for the traditional multi-
þ 12  0:0463Þ ¼ 20:5252 variate capability index shows that the product does
not comply with the manufacturer specifications
USL : UY2 ¼ ð15  ð0:0007Þ þ 24  0:9989
according to the Principal Component Analysis.
þ 15  0:0463Þ ¼ 24:6576
The centers of specifications for each component Suggested approach: Commercial multivariate
are capability index
PCmY1 ¼ ð13:5  0:0771 þ 22  ð0:0323Þ Now, consider the same dataset presented in Table 7
þ 13  0:7063Þ ¼ 9:5122 but this time with a fourth quality characteristic based
on the time it takes for a coat of the paint to dry. The
PCmY2 ¼ ð13:5  ð0:0007Þ þ 22  0:9989 manufacturer has not set acceptance criteria for this
þ 13  0:0463Þ ¼ 22:5683 variable but would like to include it in the process cap-
ability analysis. Waiting times are a key factor in the
Through the principal component analysis, the glo-
paint industry, as the more it takes for paint to dry, the
bal capability index is calculated using Eq. [4] with
longer a project will take to complete, especially in large
the Cpk;Yi variation to consider a non-centered PR.
scale businesses such as construction and design.
Note that r ¼ 2 is defined by the two components
The reliability of the product in this case relies on
extracted from the PCA results.
1
the capability of the manufacturer to provide high-
MCpk ¼ ½0:1832  0:68102 ¼ 0:3532 quality paint with lower drying times between coats
but maintaining all the other product characteristics
The Cpk;Yi values were calculated with equation that define the level of quality it possesses. For the

UYi  CYi CYi  LYi drying times data, the mean and standard deviations
Cpk, Yi ¼ min ,
3r 3r were calculated as
Where r is the square root of each component l ¼ 0:42
eigenvalue and CYi are the central values of each com-
ponent, obtained by evaluating the specifications cen- r ¼ 0:14
tral values in the normalized vector function. Thus, the shape and scale parameters for the
  

 10:9758  9:5122   9:5122  8:7548  gamma distribution are determined by


Cp, Yi ¼ min  pffiffiffiffiffiffiffiffiffiffiffiffiffi ,  pffiffiffiffiffiffiffiffiffiffiffiffiffi 
3  1:8992   3  1:8992  l2
a¼ ¼ 8:98
¼ 0:1832 r2
12 R. BARRETO AND R. HERRERA

Table 11. Quality variables in latex paint manufacturing. Table 12. Process capability indices for each variable.
Yield (l=m Þ
2
Viscosity ðPa  sÞ Thickness (mils) Drying time (h) Variable Cpu Cpl Min fCpu , Cpl g
11.84 25.20 12.94 0.5014 Yield 1.4713 0.7449 0.7449
12.03 22.05 13.14 0.6078 Viscosity 0.1675 0.6004 0.1675
12.79 19.88 13.34 0.5162 Thickness 0.6168 1.9785 0.6168
12.83 21.06 13.91 0.2209
12.83 22.01 13.91 0.4198
13.00 24.05 14.09 0.2089 proposed multivariate capability index was calculated
13.18 23.11 14.28 0.4173
13.00 24.04 14.22 0.5933 using Eq. [7].
13.15 21.41 14.26 0.423  

1 1 1:3507 1
13.18
13.26
23.15
25.95
14.20
14.37
0.312
0.4283
MCpCR ¼ 0:5  U þ 0:5  U ½0:9991
3 2
12.52 21.46 13.58 0.3907
12.86 20.43 13.94 0.6455 ¼ 0:5983
13.32 25.22 14.20 0.3845
12.94 22.87 14.14 0.2263
13.23 22.75 14.51 0.4616 The overall results are presented in Table 13, indi-
13.25 24.67 14.26 0.3965
13.26 24.33 14.37 0.3739
cating the different parameters used and their value
13.14 23.62 14.24 0.2805 for each factor included in the process capabil-
14.19 21.57 15.58 0.4092 ity analysis.
12.92 22.86 14.00 0.3643
13.25 23.77 14.36 0.2648 The following section will explore the differences
12.79 21.45 13.86 0.7604 between the two multivariate capability index
13.30 22.13 13.42 0.2367
12.97 23.22 14.16 0.5065 approaches and the advantages of the pro-
13.62 24.27 14.46 0.6141 posed method.
12.45 27.25 13.39 0.5882
13.55 21.50 14.68 0.3644
12.56 22.81 13.55 0.4364
13.04 25.74 14.12 0.2605 Results and discussion
Although reliability is an increasingly demanded qual-
And ity characteristic in the current market for every
r2 industry, its evaluation is generally overlooked in
b¼ ¼ 0:05 overall process performance evaluation practices. This
l
paper focused on the proposal of a multivariate cap-
Now, consider the scenario where the manufacturer,
ability index based on a parameter modification in
based on the drying time sample presented in Table
traditional PCA-oriented capability analysis methodol-
11, promises the customer that each layer of the paint
ogies. In the first case study, the incorporation of
will dry in one hour or less. The cumulative probabil-
product reliability into the quality performance study
ity of this outcome based on the gamma distribution
of a productive system, resulted in a more stringent
density function is
measure that could guarantee not only specification
Rð1Þ ¼ f ð1; 8:98, 0:05Þ ¼ 0:9991 and design compliance but also customer satisfaction
The next step is calculating the specifications-com- and commercial value. Through the case study of a
pliance factor A. The process capability indices Cpu pharmaceutical company specialized in Levocetirizine
and Cpl were calculated for each of the variables for production in the first case study and a Water-based
which the manufacturer had set Upper and Lower paint production process in the second case study, the
Specification Limits. Table 12 summarizes this infor- traditional technique proposed by Wang was com-
mation, along with the minimum between the pared with the proposed multivariate index. The trad-
two measures. itional method consists of using the Principal
Applying Eq. [8], the product compliance factor for Component Analysis and its respective matrix-based
the proposed model is calculation to evaluate not only the process capability
for each quality characteristic, but also the correlation
A ¼ 1:3507
between these variables.
The weighted importance factor wi was defined as In the first case study, the result obtained for trad-
0.5 for both the product compliance with the specifi- itional PCA-based multivariate capability index pro-
cations and product reliability. However, this factor posed by Wang is MCp ¼ 0:8063: Using the process
will depend on the criteria of the manufacturer and capability analysis framework proposed by Tsai and
the priorities of the process performance evaluation. Chen, this result shows that the Levocetirizine manu-
Once these parameters have been established, the facturing process is not adequate and therefore, the
QUALITY ENGINEERING 13

Table 13. Reliability analysis multivariate capability index, Case II.


Form Scale Time (h)
Parameters 8.98 0.05 1
RðtÞ 0.9991
Weight Specifications compliance ðw1 Þ Product reliability ðw2 Þ
0.5 0.5
Reliability capability index MCpCR 0.5983

products will not comply with technical specifications. 0.5 hours as opposed to one hour, the probability of
On the other hand, the reliability multivariate capabil- the product complying with this expectation would be
ity index highlights this problem in a different light. approximately 82% and the multivariate index value
In this case, MCpR ¼ 0:6317 which, according to Tsai would drop to 0.2281. This provides an opportunity
and Chen, indicates a poor process performance and to explore different scenarios for this and other reli-
defective final products. Additionally, this index con- ability-related quality characteristics in order to ana-
tains a product reliability measure that is weighted lyze the performance of the product under each set of
alongside the specification compliance value. This is conditions. If this characteristic were to be included
an advantage because it prompts decision-makers to in the traditional method, utilizing the PCA method-
develop production plans or service agreements that ology along with the other quality characteristics, the
meet technical specifications while also maintaining difference in data distributions could lead to a higher
external stakeholder engagement. level of inaccuracy in the analysis. More importantly,
The second case study, based on the observations the proposed model has the advantages of using a
collected in a water-based paint manufacturing pro- time-based model and a dedicated method for these
cess, highlights the opportunities for analysis that the type of quality characteristics even without defined
proposed model presents. The PCA-based Index technical specifications.
shows that the process has poor capability perform- The importance of this investigation relies on the
ance with an estimated value of MCp ¼ 0:3532, which replicability of the suggested approach, which can be
means that the process fails to produce units that used for any product where reliability is a quality cri-
comply with the technical specifications set by the terion. Thus, this paper constitutes a valuable contri-
manufacturer, according to Tsai and Chen parameters. bution to the business sector and the academic
The proposed model, conversely, resulted in a capabil- community, especially since this criterion has yet to
ity index with a value of MCpR ¼ 0:5983: While both be considered in traditional multivariate analysis
results show that the process is underperforming, the methods. Furthermore, the proposal exposes an inte-
proposed model offers a higher value based on the gral conception of what quality constitutes, as it
consideration of the nuances of manufacturer prom- allows the assessment of other process or service qual-
ises and product reliability, which is presented ity dimensions. In doing so, it becomes possible to
through the analysis of the drying times of the paint. analyze process performance in a more pre-
Following a gamma distribution model, the drying cise manner.
times of the paint are a feature of this specific product
that can lead to a competitive advantage for the
Future work: Multivariate capability analysis
manufacturer. The ability to promise shorter drying
model assumptions
times while maintaining compliance with the specifi-
cations and overall high-quality performance could Multivariate Capability Analysis has become a preva-
potentially put this product over others of a similar lent technique for manufacturers in the past few deca-
kind in the market. The proposed model explores this des, as researchers propose new methods for
possibility by including it in the reliability factor and analyzing process capability in multivariate settings. It
studying the scenario where the paint takes less than has been stablished that the majority of manufacturing
an hour per coat to dry. Based on the time observa- products possess more than one quality characteristic,
tions, there is a probability of 99% that this event will hence the increased interest in statistical tools to study
happen, leading to a higher reliability factor and over- these variables and their correlation, which is a key
all better performance of the manufacturing process. factor in process capability assessment (Alatefi,
The proposed model also presents the opportunity Ahmad, and Alkahtani 2019). One of the most rele-
of studying this quality characteristic with a stricter vant multivariate capability analysis tools are the
approach. If the manufacturer promise were to be Multivariate Process Capability Indices. Many authors
14 R. BARRETO AND R. HERRERA

have proposed MPCIs that take advantage of the var- Lastly, the presence of autocorrelation in the data,
iety of quality characteristics in real-life products and offer the opportunity of adjusting the parameters of
exploit the correlation among them, in order to offer the proposed model to include a scenario where auto-
new perspectives about process capability to support correlation is prominent and could affect the global
decision making. results of the capability analysis. Pan and Huang
Despite the great advantages of using MPCIs, the (2015) propose the modification of the variance-
proposed methods so far work under three main covariance matrix in the traditional methods and cre-
assumptions regarding the behavior of the observa- ating a new correction factor that considers a function
tions: normality in the data distribution, in-control between process mean and the process target vector.
process performance and mutual independence in the In conclusion, based on the considerations presented,
data. According to Sun, Shengxian, and Zhihui (2010), this research could lead to two new proposals of
there are varying levels of relevance for each of these Multivariate Capability Index that include product
assumptions in the models presented by different reliability in data that is not necessarily normally dis-
authors. The normality in the data is often regarded tributed and where the assumption of autocorrelation
as the more important condition to be met in any is taken into account with a proposed development of
study and therefore, there are multiple ways to test a new correction factor.
the fit to normal distribution. As a result, normality
tests are included in almost every MPCI investigation About the authors
published. The assumption of in-control process is Rister Barreto is an Industrial Engineering undergraduate
similarly regarded as an important feature and many and Junior Researcher, currently employed in the logistics
researchers make sure that the observations meet this and data quality field. Barranquilla, Colombia. His e-mail is
condition with several tests, such as Control Charts rjbarreto@mail.uniatlantico.edu.co
and Standard Deviation analysis. Autocorrelation, Roberto Herrera is a Professor of Statistics in the
Engineering Faculty of Universidad del Atlantico and a
conversely, is rarely studied and often disregarded in Doctor in Statistics. Barranquilla, Colombia. His e-mail is
most PCI proposals, despite having relevant effects on robertojoseherrera@gmail.com
the results as explained by Guevara and
Vargas (2007). ORCID
The content of this study sustains these assumptions Rister Barreto http://orcid.org/0000-0003-3408-4612
in the model proposition, however, there are many Herrera Acosta Roberto http://orcid.org/0000-0003-
opportunities to explore in future research with the 4855-4830
Commercial Multivariate Capability Index and the reli-
ability factor. Alatefi, Ahmad, and Alkahtani (2019) References
explain that there are many authors that have proposed
Alatefi, M., S. Ahmad, and M. Alkahtani. 2019.
MPCIs that deal with non-normality in the data.
Performance evaluation using multivariate non-normal
Nevertheless, there are some limitations to the pro- process capability. Processes 7 (11):833–57. doi:10.3390/
posed indices in this category, such as complex compu- pr7110833.
tations to find the PCI in non-normal data, depending Blischke, W. R., and D. N. Murthy. 2000. Reliability. New
on some level of normality, dealing with a limited quan- York: Wiley.
Bothe, D. 1991. A capability study for an entire product.
tity of quality variables for the PCI to be accurate, ASQC Quality Control Transations :921–5.
restricted to only a few specific cases of non-normality Castagliola, P., P. Maravelakis, S. Psarakis, and K.
and, consequently, the difficulty in the replicability of V€annman. 2009. Monitoring capability indices using run
these models. Therefore, data transformations to adhere rules. Journal of Quality in Maintenance Engineering 15
to a normal distribution for non-normal data are an (4):358–70. doi:10.1108/13552510910997733.
Chen, K. S., W. L. Pearn, and P. C. Lin. 2003. Capability
important research opportunity. Once non-normality measures for processes with multiple characteristics.
has been stablished, there are various ways to meet this Quality and Reliability Engineering International 19 (2):
condition transforming the data using methods such as 101–10. doi:10.1002/qre.513.
the Root Model, the Box Cox transformation, and the De-Felipe, D., and Benedito, E. 2017. A review of univariate
Johnson Model. These methods are relatively simple, and multivariate process capability indices. International
Journal of Advance Manufacturing Technology 92 (5):
only requiring the application of simple calculations in 1687–1705. doi:10.1007/s00170-017-0273-6.
the data and the adjustment of technical specifications Garcıa, T., M. Vasquez, G. y Ramırez, and J. Garcıa. 2007.
for the transformed data. Extension Multivariante del Indice de Capacidad Real de
QUALITY ENGINEERING 15

Procesos [Multivariate Extension of the Process Actual tool. International Journal of Research Engineering &
Capability Index. Revista Ingenierıa UC 14 (3):86–91. Applied Sciences 2 (2):1211–7.
Guevara, R, and J. Vargas. 2016. Evaluation of process cap- Steiner, S., A. Bovas, and J. MacKay. 1998. Understanding
ability in multivariate nonlinear profiles. Journal of process capability indices. Ontario, Canada: Department
Statistical Computation and Simulation 86 (12):2411–28. of Statistics and Actuarial Science-University of
doi:10.1080/00949655.2015.1112391. Waterloo.
Guevara, R., and J. Vargas. 2007. Comparison of process Sun, J., W. Shengxian, and F. Zhihui. 2010. The effect of
capability indices under autocorrelated data. Revista autocorrelated data on process capability index Cpk based
Colombia de Estadıstica 20 (2):301–16. on AR (1) model. Beijing: Department of Management
Herrera, R., Hernandez, K., Figueroa, E., and De La Ossa, Science and Engineering, Tsinghua University.
O. 2018. Application of multivariate statistical control to Taam, W., P. Subbaiah, and J. W. Liddy. 1993. A note on
measure the process capability of compression springs in multivariate capability indices. Journal of Applied
stainless steel. Prospect 16 (2):49–58.. Statistics 20 (3):339–51. doi:10.1080/02664769300000035.
Jackson, J. E. 1980. Principal component and factor analysis: Tsai, C., and C. Chen. 2006. Making decision to evaluate pro-
Part 1-Principal components. Journal of Quality cess capability index Cp with fuzzy numbers. The
International Journal of Advanced Manufacturing
Technology 12 (4):201–13. doi:10.1080/00224065.1980.
Technology 30 (3–4):334–9. doi:10.1007/s00170-005-0052-7.
11980967.
Varshini, K, Sudendar, S, Arjun, Kumaran, Abitha, S,
Murthy, D. N. 2007. Product reliability and warranty: An
Barathi, Priya, Marimuthu, Muthuvel. 2018. National
overview and future research. Production 17 (3):426–34.
conference on machine learning & artificial intelligence,
doi:10.1590/S0103-65132007000300003.
Coimbatore Institute of Technology, India, August 27th
Murthy, D. N., and R. Jiang. 2009. Impact of quality varia-
and 28th.
tions on product reliability. Reliability Engineering & Wang, F. K., and J. Chen. 1998. Capability index using
System Safety 94 (2):490–6. doi:10.1016/j.ress.2008.05.009. principal component analysis. Quality Engineering 11 (1):
Pan, J., and W. Huang. 2015. Developing new multivariate 21–7. doi:10.1080/08982119808919208.
process capability indices for autocorrelated data. Quality Wang, F. K., and T. C. Du. 2000. Using principal compo-
and Reliability Engineering International 31 (3):431–44. nent analysis in process performance for multivariate
doi:10.1002/qre.1603. date. Omega 28 (2):185–94. doi:10.1016/S0305-
Shariai, H, and Abdollahzadeh, M. A new multivariate pro- 0483(99)00036-5.
cess capability vector. Quality Engineering 21 (3):290–9. Wierda, S. J. 1994. Multivariate statistical process control -
doi:10.1080/08982110902873605. Recent results and directions for future research.
Shinde, R. L., and K. G. Khadse. 2009. Multivariate process Statistica Neerlandica 48 (2):147–68. doi:10.1111/j.1467-
capability using principal component analysis. Quality 9574.1994.tb01439.x.
and Reliability Engineering International 25 (1):69–77. Wooluru, Y., D. R. Swamy, and P. Nagesh. 2014. The pro-
doi:10.1002/qre.954. cess capability analysis – A tool for process performance
Shinde, J. H., and R. S. Katikar. 2012. Importance of process measures and metrics – A case study. International
capability and process performance indices in machine Journal for Quality Research 8 (3):399–416.

You might also like