Statistical Process Control in Big Data Environment

Darja Noskieviová Tereza Smajdorová Eva Tyleková

Department of Quality Management Department of Quality Management Department of Quality Management
VŠB – Technical university of Ostrava VŠB – Technical university of Ostrava VŠB – Technical university of Ostrava
Ostrava, Czech Republic Ostrava, Czech Republic Ostrava, Czech Republic
[Link] [Link] [Link]

Abstract—Big data analysis tools are an inevitable part of II. NEW CHALLENGES IN COLLECTING, ANALYSING AND
instruments and methods for monitoring and predicting the INTERPRETING DATA
longitudinal performance of the processes in the production
systems of the future, based on the deep automatization and A. Control Charts for High Yield Processes
overall digitalization. From this point of view statistical When products are classified as conforming or
process control (SPC) will continue to be very effective nonconforming, the proportion of nonconforming units in a
method for meeting these goals. But there must be done some process is obviously monitored by p-chart (or np-chart).
modifications. This paper deals with such possible Due to modern manufacturing technologies and concepts,
modifications of SPC. In the first part of the paper the stress many processes are of such high quality that the fraction
is put on various methods that can be integrated into SPC to
nonconforming or the probability of observing a
meet new challenges in collecting, analysing and interpreting
nonconforming unit is very small and classical Shewhart
data (control charts for high yield processes, multivariable
approaches, profile monitoring, data mining tools including
CC (p-chart) and (np-chart) has been failing [1]. For these
machine learning methods, nonparametric control charts). conditions several alternatives based on the Bernoulli
SW for the selected discussed methods is also mentioned. process have been developed. In the Bernoulli process,
The second part of the paper is devoted to the non- random values xi = 0 or xi = 1, i = 1, 2, … express whether
parametric methods of SPC and to the methodology of their an inspected item is conforming or nonconforming,
practical application. respectively. The number of units Yi being inspected until
the r-th nonconforming one has been observed (r  1).
Keywords— big data, statistical process control, Either the geometric (for r = 1) or the more general
automatization, digitalization negative binomial distribution of Yi are assumed. In [2]
authors consider four groups of methods for monitoring the
I. INTRODUCTION fraction nonconforming: Shewhart-type charts based on the
Big data analysis methods represent very important geometric distribution, Shewhart-type charts based on the
instruments for monitoring and predicting the process negative binomial distribution, CUSUM charts and
performance in the future production systems that will be EWMA charts also based on the geometric or the negative
more and more based on the deep automatization and binomial distribution. The CCC chart representing the first
digitalization. For this reason statistical process control group was designed by Calvin, see [3]. CCC means the
(SPC) will remain very effective method for meeting these cumulative count of conforming units till occurrence of
goals. But some modifications will have to be done. nonconforming unit, but more frequently the variable Y
Processes of manufacturing systems based on the that is monitored includes also the immediately following
principles and technologies of Industry 4.0 are supposed nonconforming unit. To improve the sensitivity in
to have high quality level. For that reason SPC for high detecting small upward shifts in the fraction
yield processes can be very effective way how to monitor nonconforming, the CCC-r chart for r ≥ 2 has been
such processes [1, 5]. In addition with the growing developed [4, 5, 6]. In [7] there is constructed an economic
complexity of these modern manufacturing processes the model to find the best value of r. Also in [8] the
necessity to search and monitor relations between various development of semi-economic model for setting r value
output variables and between output variables and input and its practical application can be found. Cumulative sum
variables influencing them will be stronger. It opens wider (CUSUM) charts use information from all the prior
possibilities to practically apply multivariable approaches observations and are considered an efficient alternative to
[11, 12, 16]. Profile monitoring also looks to be very the Shewhart chart when small process shifts are of
promising for monitoring and analysis of processes in the interest. The corresponding CUSUM charts for high yield
frame of modern manufacturing systems [17]. Trends in processes are called the Bernoulli CUSUM or and
automation and digitalization ask for on-line process geometric CUSUM (or CCC CUSUM), respectively (see
monitoring and predictive methods to be able to control for instance [9]). Application of CCC CUSUM chart
them. Various methods of data mining including machine can be found in [6]. Exponentially weighted moving
learning show a big potential to support solving such tasks average (EWMA) charts have similar efficiency as the
[22]. CUSUM charts. The EWMA charts for the Bernoulli and
The advantages of nonparametric methods can also be geometric distribution are studied for instance in [10].
used for effective monitoring of modern manufacturing
processes. The second part of the paper is devoted just to B. Multivariable Approaches
the non-parametric methods of SPC and to the In many industrial applications, the output of a process
methodology of their practical application [24, 25]. is characterized by p variables that are measured
simultaneously. Process monitoring problems in which
several variables are of interest are called “multivariate

978-1-7281-1951-9/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: UNIVERSITY OF WESTERN ONTARIO. Downloaded on May 26,2021 at [Link] UTC from IEEE Xplore. Restrictions apply.
quality control (or process monitoring)” problems. expressed using a simple linear model [21]. Except for
Independent variables can be charted individually, but if two-dimensional shapes, the profile monitoring might be
the variables are correlated, a multivariate chart is needed used to formation three-dimensional surfaces or
to determine whether the process is in control. geometrical profiles like flatness, roundness or
For example [11] considered a machining operation for cylindricity. Profile monitoring has numerous possibilities
valve seat inserts where the variables of interest are outside of application, for example in engineering, automotive
diameter, width, seat angle and seat concentricity. industry, aviation industry, but also in health care.
Although monitoring could be achieved by using a
univariate CC for each of the characteristics, the possible D. Data Mining Tools Including Machine Learning
Methods
correlations between variables is not utilized in this
approach. The use of procedures that capture the Data mining might be defined as automatic or
information available in bivariate (or multivariate) data on semiautomatic process of discovering pattern in data [22].
the same chart can provide richer interpretations than Patterns hidden in the data are acquired by generalizing
individual univariate charts. typical common features of data. Using relevant and
In addition [12] showed that the use of simultaneous meaningful patterns gained through data mining helps to
univariate CC could give misleading results even when the make objective decision making and might lead to
correlation is zero. business growth or obtaining economic advantage. Data
Three of the most popular multivariate quality control warehouses containing recorded big data are used as a
statistics are Hotelling’s T2 , the MEWMA (Multivariate source of data for data mining. Computational machines
exponentially-weighted moving average), MCUSUM need to learn, according to enough quality examples, how
(Multivariate Cumulative Sum). Multivariate charts like to search for these patterns. A field of training the devices
Hotelling’s T2, MEWMA, MCUSUM, take this correlation for data mining is called as machine learning. The ways in
into account in monitoring the mean vector or variance- which machines learn to search pattern among data
covariance matrix. Multivariate charts are less popular than include supervised learning, unsupervised learning and
univariate CC. Besides the relative difficulty involved in semisupervised learning. In the case of supervised
their computation, there are some additional difficulties learning, there are training examples available, where the
which diminish their appeal: " Unlike the univariate case, pattern assignment is known, so machines train the way of
the scale of the values displayed on the multivariate chart detecting pattern on these examples. The disadvantage of
is not related to the scales of any of the monitored this method is a necessity of knowing right assignment of
variables " Once an out-of-control signal is given by a examples to the patterns however it is also the most
multivariate chart, it may be difficult to determine which effective learning method. In unsupervised learning, the
variable caused this signal. More complicated operations assignment of examples to patterns is not known, which
are required to determine the origin of the signals [11, 12, means there is no possible feedback to determine in which
13, 14, 15, 16]. cases a algorithm failed. The machine is able to create
clusters of similar attributes and use these clusters as
C. Profile Monitoring classification classes or generalized patterns.
The practical applications of statistical process control Unsupervised learning represents a less effective way than
generally work on the assumption that the quality of supervised learning. The compromise between supervised
process or product might be expressed through and unsupervised learning is semisupervised learning. In
distribution of a single particular quality characteristic in case of these method, the machine is provided with only a
an univariate CC or through multivariate distribution of limited number of training examples for which a right
vector in a case of several correlated quality assignment is known. Therefore, the algorithm gradually
characteristics. Nevertheless, it appears that in some cases compares unknown examples to known ones and, based
the quality of process or product might be better described on their similarity, assigns the unknown ones to the most
by a function so-called profile [17]. A profile represents a similar pattern. Such method is more effective than simple
functional relationship between dependent variable and at unsupervised learning. The techniques used in data mining
least one independent variable. This function or profile is include association, clustering, predictions, classification
detected and monitored over time. The prerequisite of or sequential patterns [23]. Association technique seeks to
using each profile is the measuring of n values of the find relationship among items that typically occur
dependent variables together with corresponding values of together. Afterwards on the basis of these formed
one or more explanatory or independent variables [18]. association rules, the other items are assigned. Searching
Profile monitoring is used to determine and control the for association rules among datasets represents a
stability of functional relationship between variables over descriptive method of data mining. Clustering technique
time. Methodology of profile monitoring is based on the identifies an inherent data grouping pattern based on their
fact that the observed values of the quality characteristics known features. A mutual similarity is sought between
can be recorded as a set of data points that can form a data and if the data meet a certain degree of similarity, a
curve (profile). So unlike monitoring single measurement cluster is created representing the desired generalization
on each product or unit, a set of values is observed in a (pattern). Clustering is predictive method that represents
range that creates a profile when plotted on a graph [19]. the unsupervised learning and its application does not
Numerous types of models might be used to describe the require a learning set. Sequential patterns seek patterns
profile – from simple linear regression models, nonlinear that are similar over time, for example within a business
regression models, multiple regression models, period. Classification technique is supervised method that
polynomial regression models, nonparametric regression requires a learning set. Based on the learning set that
to mixed models and wavelets models [20]. Of the many contains items with known classification, the method of
possible regression models, the profile is most often classification is learned and might be further applied to

Authorized licensed use limited to: UNIVERSITY OF WESTERN ONTARIO. Downloaded on May 26,2021 at [Link] UTC from IEEE Xplore. Restrictions apply.
other unclassified data. Therefore classification is developed, which are not affected by breaking assumptions
predictive method. Some frequently used methods for about the data [24, 25].
classification are decision trees, support vector machine, In practice, nonparametric methods are not known,
multivariate adaptive regression splines or neural which is due, among other things, to the lack of
networks. Suitable software for applying these methods is comprehensive guidance for their application. For this
Statistica by Tibco. reason the deep and complex analysis of performance of
the selected non-parametric control charts (NPCC) were
done and the methodology of application of NPCC in
E. Nonparametric Control Charts practice was proposed [32]. An overview of the advantages
Classical parametric statistical methods are based on and disadvantages of the analysed NPCC is in Tab. I.
several basic assumptions about data (normality, These are NPCC for monitoring position SSCC [26], NP-
independence, constant mean and variance). Unfortunately, EWMA [27], NP-CUSUM [28] and PM [29] and NPCC
these assumptions are not always fulfilled in practice. for monitoring variability Mood [30] and MAD [31]. In
When we apply parametric methods to such data it is not Tab. II there are formulas for computation of the applied
guaranteed that they will provide the right results. For control statistic, lower control limit (LCL) and upper
these cases, reliable nonparametric statistical methods were control limit (UCL) for selected analysed NPCC.

TABLE I. ADVANTAGES AND DISADVANTAGES OF NON-PARAMETRIC CONTROL CHARTS. SOURCE: OWN RESEARCH

Nonparametric Control Chart (NPCC) Advantages Disadvantages

• The easiest NPCC • Not too strong nonparametric

Shewhart sign CC (SSCC) statistics
• Easy calculations
• More complicated calculations of
NP – EWMA (NPEWMA) • NPCC with memory control limits

• It quickly reveals small and middle • More complicated calculations of

NP – CUSUM (NPCUSUM) changes in the process needed statistics
• NPCC with memory
• Statistics of progressive mean is
considered more powerful than classic
Progressive mean (PM) Shewhart CC or CUSUM and EWMA • More complicated calculations
• More effective in detecting small shifts in
the process
• NPCC with memory
• It's the most used NPCC when normality • More complicated calculations
Median absolute deviation (MAD) assumption fails • This is not a CC with memory
• It works with robust statistics via median

• Mood's statistics is considered one of the • More complicated calculations

NPCC based on Mood statistic (Mood) • This is not a CC with memory
strongest nonparametric statistics

TABLE II. FORMULAS FOR COMPUTATION OF CONTROL STATISTIC, LCL AND UCL FOR SELECTED ANALYSED NPCC

Nonparametric Control Chart (NPCC) Statistics CL, LCL, UCL

ͳǡ ‫ݔ‬௜௝ ൐ ߠ଴ ܷ‫ ܮܥ‬ൌ ܿ
‫݊݃݅ݏ‬ሺ‫ݔ‬ሻ ൌ ‫ݔ‬௜௝ െ ߠ଴ ൌ ቐ Ͳǡ ‫ݔ‬௜௝ ൌ ߠ଴ ‫ ܮܥ‬ൌ Ͳ
Shewhart sign CC (SSCC) െͳǡ ‫ݔ‬௜௝ ൏ ߠ଴
‫ ܮܥܮ‬ൌ െܿ
Sܰ௜ ൌ σ௡௝ୀଵ൫‫ݔ‬௜௝ െ ߠ଴ ൯
c = 2t – n

‫ܹܯ‬௜ െ ‫ܧ‬଴ ሺ‫ܹܯ‬௜ ሻ ߂

ܵ‫ܹܯ‬௜ ൌ ݇ൌ
ඥ‫ݎܽݒ‬଴ሺ‫ܹܯ‬௜ ሻ ʹ
NP – CUSUM (NPCUSUM)
ܵ௜ ൌ ƒšሼͲ Ǣ ܵ௜ିଵ ൅ ܵ‫ܹܯ‬௜ െ ݇ሽ ‫ ܪ‬ൌ ݇ ‫ߪ ڄ‬,
ܵ଴ ൌ Ͳ

Simulations were done for in-control and out-of-control

III. METHODOLOGY OF THE PRACTICAL APPLICATION OF process. In-control process means statistically stable
NONPARAMETRIC METHODS OF SPC process when only random causes of variation have an
The aim of this part is to describe the methodology of influence on it. The out-of-control process means that the
application of NPCC in practice. This methodology is process is statistically unstable i.e. it is influenced by
based on the results of very complex and extensive random and also assignable causes of variation.
simulations of the NPCC performance [32].

Authorized licensed use limited to: UNIVERSITY OF WESTERN ONTARIO. Downloaded on May 26,2021 at [Link] UTC from IEEE Xplore. Restrictions apply.
 The number of the subgroups m was equal to 20, 100 performed but only for 20 x 5 combination and selected
and 300, and the size of the subgroup n was chosen to be 5 types of distribution. The deviation occurred before it was
and 10. 10,000 replicates were performed. The indicators signalized by a point outside the limit. Subsequently, the
Median Run Length MRL(0) and five percent quantile (x5) process was “intervened" and next subgroup was devoid
were used to evaluate the performance of particular NPCC of deviation. Then the deviation reappeared. Only MRL()
for in-control processes. MRL(0) and x5 are based on the and x95 indicators were used to assess NPCC performance
Run Length for in-control process and it is referred to as for out-of-control process. Minimum values of these
RL(0). RL(0) is determined as the number of points performance indicators are required. The calculation of
recorded in control chart that lie within the control limits these indicators is based on RL(). This is determined as
between 2 points outside these limits and maximum the number of points that are recorded in the control chart
values of these performance indicators are required. from the moment the change in the process occurred until
The simulations for the out-of-control process were the change was signalized in the form of the point out of
realized in the same way as experiments for in-control control limit.
process but in addition deviations of different sizes  were For simulations of various violations of data
inserted into the data files and thus the out-of-control assumptions the following probability distributions were
process was simulated. First, isolated deviations of 1.5; 2 used for both in-control and out-of-control processes (see
and 3 were inserted. The isolated deviation was inserted Tab. III).
into 30th, 5030th and 9930th repeating. Subsequently, a Based on the previous analysis it can be determined
simulated persistent deviation of 1.5; 2 and 3 was also which nonparametric control chart is the most suitable for

TABLE III. OVERVIEW OF APPLIED TYPES OF PROBABILITY DISTRI BUTIONS (SOURCE: OWN RESEARCH)

Distribution/Type of breach of
Distribution no. Distribution parameters Unfulfilled assumption of data
assumption

1. Normal distribution N (0,1) Neither

Student´s distribution/ greater

2. kurtosis than the normal t3 Data normality
distribution

Uniform distribution/ less

3. kurtosis than the normal ܴ൫െξ͵Ǣ ξ͵൯ Data normality
distribution

Pearson distribution/ skewed

4. ߯ଷଶ Data normality
distribution

5. Mixed distribution 50% N(0,1) + 50% N(2,1) Constant mean

6. Mixed distribution 50% N(0,1) + 50% N(0,4) Constant variance

Time series/ Autocorrelated

7. AR (1): xi – 0.5xi+ai Independence of data
data

a given data assumption violation. The results for NP –

The results also showed that for the in-control process,
CUSUM and SSCC when the process is in-control are
NPCCs perform very well, but for the out-of-control
shown in Tab. IV. These control charts were determined
process with an isolated deviation their performance is
as quite robust (universal) control charts and from them
surprisingly poor, especially for small process changes. It
the most robust control chart (i.e. the one that has the most
can be said that when the size of the subgroup grows, the
stable performance indicator values over all probability
performance indicators improve. Worse performance
distributions) was set. Our analysis showed that the most
indicators results for the out-of-control process with
universal control chart is SSCC, where the performance
isolated deviation may be due to the reason that
indicator values were the most stable for different
nonparametric control chart perceive the isolated
deviations from data assumptions [32].
deviation as a random deviation against which they are
TABLE IV. PERFORMANCE ANALYSIS RESULTS –
robust [32].
IN-CONTROL PROCESS (SOURCE: OWN RESEARCH) The analysis of the MRL() and x95 performance
indicators showed that the persistent deviation is detected
NP - CUSUM SSCC by NPCC very quickly. The results showed that NP-
· Nonconstant mean value · Autocorrelation data EWMA is the most powerful, as with the isolated
· Nonconstant variance · Nonconstant mean value deviation. Other NPCCs that have good x95 results are NP-
· Greater kurtosis than · Skewed distribution CUSUM and SSCC [32].
normal distribution · Greater kurtosis than normal
· Skewed distribution distribution
The proposed methodology for practical application of
NPCC has been based on these own simulation results
[32]. The designed methodology has four phases: 1. -

Authorized licensed use limited to: UNIVERSITY OF WESTERN ONTARIO. Downloaded on May 26,2021 at [Link] UTC from IEEE Xplore. Restrictions apply.
preparatory phase, 2. - phase of data collection and classic Shewhart control chart, for the nonparametric ones
analysis, 3. - phase of selection of suitable control chart is valid that if all points are within the control limits, then
and 4. - phase of evaluation of the process statistical the process can be considered in-control. However, if any
stability. of the points in the control chart lies out of the control
Phase 1 includes determination of the quality limit, it is a sign that the process is out-of-control. That is,
characteristic or process parameter, which we want to the process is affected by an assignable cause of variability
regulate, selection of the suitable control point, choice of that needs to be identified, then its root cause must be
the method of data collection and recording (sample size found, removed and the statistical stability assessment
and the frequency of sampling). Phase 2 is an important must be performed again.
part of SPC, because here we verify whether the basic data The proposed methodology was verified on real data
assumptions are met and therefore whether the classic from an organization operating in the automotive industry.
Shewhart control chart can be used, or whether the data The application of this methodology led to the
assumptions are not met and it will be preferable to use one improvement of the verified process and the entire system
of the nonparametric control charts. Verification of data of statistical process control in the organization. [32].
assumptions can be performed using various statistical
tests or graphical methods. It is always advisable to IV. CONCLUSIONS
combine multiple tests with a graphical method. The phase The paper has dealt with various methods suitable for
3 is devoted to the selection of suitable control chart. If all monitoring and control of processes in modern
data assumptions are met, the classic Shewhart control manufacturing systems. The research has shown that they
chart can be applied. The method of selecting a suitable have a big potential for solving these tasks responding to
Shewhart control chart, its construction and evaluation is various specific aspects of the Industry 4.0.
described in detail in publications about SPC. The problem SPC for high yield processes can be very effective
occurs when some of the data assumption is not met. In way how to monitor processes in modern manufacturing
that case, the use of the classic Shewhart control chart systems as such processes are supposed to have very high
could lead to misleading results and, at worst, a quality level. The growing complexity of these processes
deterioration of the process. Therefore, the application of opens wider possibilities to practically apply multivariable
nonparametric control charts is a possible alternative for approaches enabling to monitor and control relations
these cases. The process of selection of suitable NPCC between various output variables and between output
control chart is depicted in Fig. 1. variables and input variables influencing them. Profile
monitoring evinces high ability for the process monitoring
and analysis in such situation. Various methods of data
mining including machine learning show capabilities to
support possibility for prediction of the processes
behaviour which is required by the on-line monitoring and
control that result from the trends in automatization and
digitalization. Nonparametric methods can also be used
for effective monitoring and analysing modern
manufacturing processes. For that reason the second part
of the paper was devoted just to the non-parametric
methods of SPC and to the methodology of their practical
application. The proposed methodology was verified on
the real data from the processes at the automotive
supplier.
ACKNOWLEDGMENT
The work was supported by the European Regional
Development Fund ERDF project “Research Platform
focused on Industry 4.0 and Robotics in the Ostrava
Agglomeration” CZ.02.1.01/0.0/0.0/17_049/0008425 in
the frame of the Operational Programme Research,
Development and Education and the specific university
research of the Ministry of Education, Youth and Sports of
the Czech Republic No. SP2020/109.

REFERENCES
[1] D. Noskieviová, E. Jarošová, and K. Brodecká, “Application of
Bernoulli process-based charts to electronic
assembly”, International Journal of Engineering Business
Management, InTech, 2015, pp. 1-10.
[2] J. l. Szarka, and W. H. Woodall, “A review and perspective on
Fig. 1. Flowchart for application of non-parametric control charts surveillance of Bernoulli processes”, Qual. Reliab. Eng. Int., No.
into practice. Source: own research 27, 2011, pp. 735-752.
[3] T. Calvin, “Quality control techniques for zero defects”, IEEE
During the last proposed phase the evaluation of the Transactions on Components Hybrids and Manufacturing
process statistical stability must be done. As with the Technology, vol. 6, 1983, pp. 323-328.
[4] W. Xie, M. Xie, and T. N. Goh, “A Shewhart-like charting

Authorized licensed use limited to: UNIVERSITY OF WESTERN ONTARIO. Downloaded on May 26,2021 at [Link] UTC from IEEE Xplore. Restrictions apply.
 technique for high yield processes”, Qual. Reliab. Eng. Int., No. 11, Engineering International [online]. vpol. 31, No. 1, 2015, pp. 3-13,
1995, pp. 189-196. Available from: [Link]
[5] E. Jarošová, and D. Noskieviová, “Comparative statistical [28] D. Wang, L. Zhang and Q. Xiong, „A non parametric CUSUM
analysis of selected control charts for highly capable processes”, E control chart based on the Mann– Whitney statistic“.
+ M Ekonomie and Management , vol. XII, No. 2, 2019, pp. 68 – Communications in Statistics - Theory and Methods [online]. 2016,
82. vol. 46, No. 10, 2016 Available from:
[6] E. Jarošová, E., and D. Noskieviová, “Advanced methods of [Link]
statistical process control, “Pokroilejší metody statistické regulace 3314.
procesu”, Grada Publishing, 2015.
[7] H. Ohta, E. Kusukawa, and A. Rahim “A CCC-r Chart for High- [29] S. Škapa, “Proposal of non-Gaussian characteristics production
Yield Processes”, Qual. Reliab. Eng. Int. No. 17, 2001, pp. 439- process control”, “Návrh ízení výrobního procesu ne-
446. Gaussovského rozložení: Doctoral Thesis. Brno: Vysoké uení
technické, 2002.
[8] K. Brodecká K. “Application of selected statistical methods at
conditions of high yield processes”, “Aplikace vybraných [30] H. Murakami and T. Matsuki. “A nonparametric control chart
statistických metod v podmínkách vysoce zpsobilých proces”, based on the Mood statistic for dispersion”, The International
Doctoral Thesis, Ostrava, VŠB-TU Ostrava, 2013. Journal of Advanced Manufacturing Technology [online]. 2010,
[9] T. C. Chang,, and F. F. Gan, “CumulativesSum charts for high vol. 49, No.5, pp. 440 – 449, Available from:
yield processes”. Stat. Sinica No. 11, 2001, pp. 791–805. [Link]
[10] T. Kotani, E. Kusukawa, and H. Ohta, (2005) Exponentially [31] K.S. Adekeye and P.I. Azubuike. “Derivation of the limits for
weighted moving average chart for high-yield processes. An control chart using the median absolute deviation for monitoring
International Journal of Industrial Engineering and Management non-normal process”. Journal of Mathematics and Statistics
Systems, vol. 4, No.1, 2005, pp.75-81. [online]. vol. 8, No. 1, 2012, pp. 515-523, Available from:
[11] D. H. Kao, “Design of a package of statistical charts for bivariate [Link]
Gaussian process control”, Doctoral Thesis, University of [32] T. Smajdorová. “Application of nonparametric methods in
Wisconsin, 1986. statistical process control”, “Využití neparametrických metod ve
[12] J. E. Jackson, “Quality control methods for two related variables”, statistické regulaci proces”, Doctoral thesis, VSB – Technical
Industrial Quality Control, No. 12, 1956, pp. 2–6. University of Ostrava, Ostrava, 2019.
[13] F. B. Alt, “Multivariate quality control.” In Encyclopedia of
Statistical Sciences, vol. 6, New York, John Wiley & Sons, 1985.
[14] H. M. Bush, P. Chongfuangprinya, V. C. P. Chen, T. Sukchotrat
and S. B. Kim. “Nonparametric multivariate control charts based
on a linkage ranking algorithm”, Quality and Reliability
Engineering International [online], 2010, vol. 26, No. 7, pp. 663-
675, Avalable from: [Link]
[15] SAS/QC® 14.2 User’s Guide: The shewhart Procedure [online].
14.2. USA: SAS Institute, 2016 [cit. 2019-12-03]. Avalable from:
[Link]
[Link]?locale=en#nameddest=qcug_shewhart_sect522.
[16] Ü. O. Firat, “Multivariate Quality Control: A Historical
Perspective”. Tildiz University, Available from:
[Link]
[17] W. H. Woodall and D. C. Montgomery, “Some current directions
in the theory and application of statistical process monitoring,”
Journal of Quality Technology, vol. 46, 2014, pp. 78-94.
[18] R. Noorossana, A. Saghaei, and A. Amiri, “Statistical analysis of
profile monitoring”. John Wiley & Sons, 2011.
[19] D. C. Montgomery, Introduction to Statistical Quality Control.
John Wiley & Sons, 2019.
[20] W. H. Woodall, “Current research on profile monitoring,”
Produção, vol. 17, 2007, pp. 420-425.
[21] M. R. Maleki, A. Amiri, and P. Castagliola, “An overview on
recent profile monitoring papers (2008–2018) based on conceptual
classification scheme,” Computers & Industrial Engineering, vol.
126, 2018, pp. 705-728.
[22] I. H. Witten, E. Frank, M. A. Hall and C. J. Pal, “Data mining:
practical machine learning tools and techniques”, Elsevier, 2017.
[23] D. L. Olson, and D. Delen, Advanced data mining techniques,
Springer, 2008.
[24] S.T. Bakir, T. Prater and S. Kiser, “A simple nonparametric
quality control chart for monitoring students’ GPAs”. SOP
Transactions on Statistics and Analysis [online]. vol. 1, 2015, pp.
8-16 [cit. 2017-09-26]. DOI: 10.15764/STSA.2015.01002. ISSN
2373843x. Available from:
[Link]
[25] J. Dempír and L. Dohnal, “Some robust methods for setting mean
and dispersion of the ste of results and their applications”, “Nkteré
robustní postupy urení stední hodnoty a rozptýlení souboru
výsledk a jejich použití”, Klin. Biochem. Metab., vol. 13, No. 3,
2005, pp. 139–144.
[26] S. T. Bakir, „Classification of distribution-free quality control
charts“, Proceedings of the Annual Meeting of the American
Statistical Association, 2001.
[27] S. L. Lu, „An extended nonparametric exponentially weighted
moving average sign control chart“, Quality and Reliability

Authorized licensed use limited to: UNIVERSITY OF WESTERN ONTARIO. Downloaded on May 26,2021 at [Link] UTC from IEEE Xplore. Restrictions apply.