A Research Study on Unsupervised Machine Learning Algorithms for Early Fault

Detection in Predictive Maintenance

Nagdev Amruthnath Tarun Gupta

Department of IEE and EDMM Department of IEE and EDMM
Western Michigan University Western Michigan University
Kalamazoo, Michigan, USA Kalamazoo, Michigan, USA
[email protected] [email protected]

Abstract— the area of predictive maintenance has taken a lot of parameters but also to collect the physical health aspects of
prominence in the last couple of years due to various reasons. the machine such as vibration, pressure, temperature,
With new algorithms and methodologies growing across acoustics, viscosity, flow rate and many as such. This
different learning methods, it has remained a challenge for information is widely used for early fault detection, fault
industries to adopt which method is fit, robust and provide identification, health assessment of the machine and predict
most accurate detection. Fault detection is one o f the critical the future state of the machine. Some of this is made possible
components of predictive maintenance; it is very much needed due to machine learning algorithms available across different
for industries to detect faults early and accurately. In a learning domains.
production environment, to minimize the cost of maintenance,
sometimes it is required to build a model with minimal or no
historical data. In such cases, unsupervised learning would be
Machine learning is a subsection of Artificial Intelligence
a better option model building. In this paper, we have chosen a
Figure 1. Machine learning can be defined a program or an
simple vibration data collected from an exhaust fan, and have algorithm that is capable of learning with minimum or no
fit different unsupervised learning algorithms such as PCA T2 additional support. Machine learning helps in solving many
statistic, Hierarchical clustering, K-Means, Fuzzy C -Means problems such as big data, vision, speech recognition, and
clustering and model -based clustering to test its accuracy, robotics [7]. Machine learning is classified into three types.
performance, and robustness. In the end, we have proposed a In supervised learning, the predictors and response variables
methodology to benchmark different algorithms and choosing are known for building the model, in unsupervised learning, ,
the final model only response variables are known, and in reinforced
learning, the agent learns actions and consequences by
Keywords- Predictive maintenance, Fault detection, interacting with the environment. In this research, the main
manufacturing, machine learning, Just in Time focus will be on unsupervised learning methodology. One of
the most commonly used approaches in unsupervised
I. INTRODUCTION learning is clustering where, response variables are grouped
The concept of predictive maintenance (PdM) was into clusters either user-defined or model based on the
proposed a few decades ago. PdM is also a subset of planned distance, model, density, class, or characteristic of that
maintenance. PdM did not gain prominence until the recent variable. For this research, vibration data has been used. Data
decade. This rapid advance is mainly due to emerging collection, feature selection, and extraction will be described
internet technologies, connected sensors, systems capable of in the later sections.
handling big data sets and realizing the need to use these
techniques. The abrupt growth can also be theorized due to
the demand for high-quality products, at the least cost and
with shortest lead time. Every year, it is estimated that U.S.
industry spends $200 billion on maintenance of plant
equipment and facilities and the result of ineffective
maintenance leads to a loss of more than $60 billion [1]. In
food and beverage industry it was estimated that failures and
downtime accounted for 18% of OEE [2]. Over the years,
different architecture, algorithms, and methodologies have
been proposed. One of the most prominent methods is
watchdog agent, a design enclosed with various machine Figure 1: Structure of Learning Methods
learning algorithms [3] [11]. Some of the other architectures
are an OSA-CBM architecture [4], SIMAP Architecture [5], All the programming in this research is performed in a
and predictive maintenance framework [6]. Emerging statistical tool called as R- Programming. R- Program is
technologies such as the Internet of things (IoT) devices have open source software and was designed by Ross Ihaka and
formed a gateway to connect to machines and its Robert Gentleman in August 1993. As of today, there are
subcomponents to not only collect the process data and its over 10,000 packages which include thousands of different
algorithms contributed by various authors for different III. FAULT DETECTION
applications.
Fault detection is one of the most critical components of
II. LITERATURE REVIEW predictive maintenance. Fault detection can be defined as a
The primary goal of PdM is to reduce the cost of a process of identifying the abnormal behavior of a
product or service and to have a competitive advantage in the subsystem. Any deviation from a standard behavior can be
market to survive. Today business analytics are embedded categorized as a failure. In this section, we will discuss
across PdM to realize the need for it and to make appropriate different algorithms such as Principle Component Analysis
decisions. Business analytics can be viewed in three different (PCA) T2 statistic, Hierarchical clustering, K- Means
prospective (i) Descriptive analytics (ii) Predictive analytics clustering, C-Means, and Model-based clustering for fault
and (iii) Prescriptive analytics [16]. Descriptive analytics is a detection and benchmark its results for vibration monitoring
process of answering questions like what happened in the data.
past? This is done by analyzing historical data and
summarizing them in charts. In maintenance, this step is A. Data Collection
performed using control charts. Predictive analytics is an Vibration data is one of the most commonly used
extension to descriptive analytics where historical data is technique to detect any abnormalities in a submachine. In
analyzed to predict the future outcomes. In maintenance, it this research paper, a vibration monitor sensor was set up on
is used predict type of failure and time to complete failure. an exhaust fan. The vibration was collected every 240
Finally, prescriptive analytics is a process of optimization to minutes for 12 days at a sampling frequency of 2048 Hz on
identify the best alternatives to minimize or maximize the both X and Y axis. From the following data, different
objective. This also answers the questions such as what can features were extracted such as peak acceleration, peak
be done? In maintenance, this can be used to optimize the velocity, turning speed, RMS Velocity, and Damage
maintenance schedules to minimize the cost of maintenance. accumulation. Figure 2 is the time series plots of the data.
In this paper, our primary focus will be on descriptive and
predictive analytics to detect the faults.

Predictive analytics has spread its applications into

various applications such as railway track maintenance,
vehicle monitoring [23], automotive subcomponents [8],
utility systems [19], computer systems, electrical grids [13],
aircraft maintenance [21], oil and gas industry,
computational finance and many more.

Fault detection is one of the concepts in predictive

maintenance which is well accepted in the industry. Early
Failure detection could potentially eliminate catastrophic
machine failures. In one of the recent research studies, this
process is classified into different methods such as Figure 2: Feature data plot
quantitative model-based methods, qualitative model-based
methods, and process history based methods [25]. In Figure 2, we can see a trend line generating closer to
index 60th observation. In this paper, we will test to see how
Principle component analysis (PCA) is one of the oldest different algorithms help in detecting this fault earlier.
and most prominent algorithms that are widely used today. It
was first invented by Karl Pearson in 1901. Since then, they
have been many hybrid approaches to PCA for fault B. Feature Selection using PCA
detection such as using Kernel PCA [17], adaptive threshold Not all features extracted provide a true correlation. If
using Exponential weight moving average for T 2 and Q right features are not selected, then a significant amount of
statistic [9], multiscale neighborhood normalization-based noise would be added to the final model and hence, reduce
multiple dynamic principal component analysis (MNN- the accuracy of the model. One of the most prominent
MDPCA) method [27], Independent Component Analysis. algorithms for that is used for dimensionality reduction is
Another common method used for fault detection is Principle component analysis. Principal component analysis
clustering method. Similar to PCA, there are various (PCA) is a mathematical algorithm that reduces the
algorithms such as neural net clustering algorithm neural
dimensionality of the data while retaining most of the
networks and subtractive clustering [28], K-means [10],
Gaussian mixture model [15], C-Means, Hierarchical variation (information) in the data set [18]. In a simple
Clustering [22], and Modified Rank Order clustering context, it is an algorithm to identify patterns in data and
(MROC) [33]. expressing such a way to showcase those similarities and
differences [29].
Algorithm: C. T2 Statistic
Step 1: Consider a data matrix X T2 Statistic is a multivariate statistical analysis. The T 2
[X]mxn statistic for the data observation x can be calculated by [12]
(1)
Where, X is the matrix, m is a row, and n is a
��
column = ∑
��
�=
Step 2: Subtract the mean from each dimension (10)
The upper confidence limit for T 2 is obtained using the F-
[�̅] − [�] distribution:
(2)
Step 3: Calculate the covariance matrix � �−
[�] � , ,∝ = , − ,∝
�−�
(3) (11)
Step 4: Calculate the eigenvectors and eigenvalues of the
covariance matrix
[�] � − � � � {�} � = { }
(4)
Step 5: Store the eigenvector in a matrix

[�] � = [ {� }{� }{� } …. . {� }]

(5)
Step 6: Store eigenvalues in a diagonal matrix

[ ����] �
(6) Figure 3: Summary of PCA
Where [Eigen] is the eigenvalues corresponding to the
principal components, and P contains the loading vectors Where n is the number of samples in the data, a is the
number of principal components, and α is the level of
Step 7: Rank eigenvalues in decreasing order and choose significance [24]. This statistic can be used to measure the
top “r” vectors to retain values against the threshold and any values above the
[ ����] ��� threshold; can be concluded as out of control data. In this
(7) case, it is going to be faulty data. The results for the
Step 8: Retain “r” eigenvectors vibration data are shown the Figure 5.
[�] �� = [ {� }{� }{� } …. . {��}]
(8) Based on the results from T2 statistic in Figure 5, we can
Step 9: Calculate the principal components [U] which is observe that the faults can be detected as early as 41
projected in data matrix observations. Hence, this early detection would help the
[�] � [�] �� = [ ] �� maintenance teams to monitor these process changes and
(9) take corrective actions accordingly.
Summary of the PCA indicates that the first two principal
components show 95.65% of variance compared to rest of D. Cluster Analysis
the components.
Clustering analysis is one of the unsupervised learning
A scree plot can be plotted for Eigenvalues versus methods. In cluster analysis, similar data are grouped into
principle components as shown in Figure 4. This plot can be different clusters. Some of the most prominent cluster
used to define the components that show significant analyses are K-Means clustering, C-Means clustering, and
variance in the data. hierarchical clustering. There are various merging principles
in hierarchical clustering. They are iterative, hierarchical,
From summary data and scree plot, we can conclude that density based, Metasearch controlled and stochastic. In this
the first two principal components present maximum paper, we will be discussing one of the commonly used
variation compared to the rest of the principal components. hierarchical clusterings.
E. Optimal number of Clusters theorize each cluster represents a normal condition, warning
In cluster analysis, we need to know the optimal number condition, and faulty condition. In the next section of cluster
of clusters that can be formed. Although we know that, we analysis, we can observe how each of the clustering
have healthy data and faulty data, identifying the number of algorithms provides the results.
optimal cluster formations in our data would help in
understanding different states in the data and representing From both the procedures shown in Figure 6 and Figure
the data more accurately. To identify the number of clusters, 7, we can identify that 3 clusters are the optimal number of
there are many procedures available such as elbow method, clusters. For fault detection, we can use three clusters and
Bayesian Inference Criterion method and nbClust package theorize each cluster represents a normal condition, warning
in R. The results for elbow method is shown in Figure 6 and condition, and faulty condition. In the next section of cluster
using nbClust [30] is shown in Figure 7. analysis, we can observe how each of the clustering
algorithms provides the results.

Figure 4: Scree plot to determine the variation between

principal components
Figure 6: Determining the optimal number of clusters
based on Elbow Method

Figure 5: T2 statistic results for training dataset and

testing dataset

From both the procedures shown in Figure 6 and Figure Figure 7: Determining the number of Clusters using
7, we can identify that 3 clusters are the optimal number of nbClust Package
clusters. For fault detection, we can use three clusters and
F. Heirarchical Clustering
Start by assigning each item to its own cluster, so that if Table 2: Fuzzy C-Means Cluster Centers with 3 clusters
you have N items, you now have N clusters, each containing 1 2
just one item. Let the distances (similarities) between the 1 1.275 -1.071
clusters equal the distances (similarities) between the items 2 -0.289 1.920
they contain [24]. 3 -9.935 -1.723

Algorithm: From Table 3 summary of K-means and C-means clustering,

Step 1: Find the closest (most similar) pair of clusters and we can observe that clusters of sizes 4, 27 and 40 are
merge them into a single cluster, so that now you formed. Observation 1 to 40 formed one cluster, 41 to 67
have one less cluster. formed second cluster and the third cluster with 68 to 71
Step 2: Compute distances (similarities) between the new observations. These results are same as hierarchical
cluster and each of the old clusters. clustering.
Step 3: Repeat steps 2 and 3 until all items are clustered into
a single cluster of size N.

In Figure 8, the cluster is formed based on the feature

data using Ward's method. Irrespective of feature data and
Principle components, the results were identical. Three
clusters were formed, where the first cluster includes
observations from 1 to 40, the second cluster includes
observations 41 to 67 and finally, the third cluster includes
observations from 68 to 71. Based on the domain
knowledge, we can represent cluster 1 as healthy dataset,
cluster 2 as warning dataset and finally cluster 3 as faulty
data set.

G. K-Means and Fuzzy C-means Clustering

K-means is one of the most common unsupervised
learning clustering algorithms. This most straightforward
algorithm’s goal is to divide the data set into pre-determined
clusters based on distance. Here, we have used Euclidian
distance. The graphical results as shown in Figure 9

C-means is a data clustering technique where each data Figure 8: Hierarchical Clustering Solution for Fault
point belongs to every cluster at some degree. Fuzzy C Identification
means was first introduced by Bezdek [14]. Fuzzy C-Means
has been applied in various applications such as agricultural,
H. Model-Based Clustering
engineering, astronomy, chemistry, geology, image analysis
[14], medical diagnosis, and shape analysis and target A Gaussian mixture model (GMM) is used for modeling
recognition [26]. The graphical results for C-Means is as data that comes from one of the several groups: the groups
shown in Figure 9 might be different from each other, but data points within
the same group can be well-modeled by a Gaussian
Summary of K-Means and C-Means Clustering distribution [20]. Gaussian finite mixture model fitted by
EM algorithm is an iterative algorithm where some initial
Table 1: Cluster Means of K-Means Algorithm random estimate starts and updates every iterate until
1 2 convergence is detected [31] [32]. Initialization can be
1 -9.665 -1.609 started based on a set of initial parameters and start E-step
or set of initial weights and proceed to M-step. This step can
2 -0.497 1.856
be either set randomly or could be chosen based on some
3 1.301 -1.092 method.

Within cluster sum of squares by cluster: Summary of Classification

[1] 16.758705 39.575966 8.823486 Mclust EVV (ellipsoidal, equal volume) model with five
(between_SS / total_SS = 90.2 %) components:
 each of three states. The first state was identified as healthy
log.likelihood n df BIC ICL state (since it was calibrated for healthy data), second state
-57.23501 71 25 -221.037 -222.0734 was identified as the warning state and finally the third state
was identified as faulty state. It would not be surprising to
obtain the following results as all these algorithms were
based on a distance measure.

Figure 9: K-Means and C-Means clustering for fault

identification Figure 10: Gaussian finite mixture model fitted by EM
algorithm Classification
The results are summarized in Table 3. The results from
Gaussian finite mixture model fitted by EM algorithm
For our final model, Gaussian finite mixture model fitted
Classification, there was a total of 5 groups of components by EM algorithm was used. Unlike providing the number of
are formed. Component 1 and two are assigned to clusters, this model identifies optimal clusters and
observation 1 to 40, component group 3 consists of
accordingly classifies the observations into groups. Here,
observation 41 to 63, component group 4 consist of the model recognized a total of 5 components. Although
observations 64 to 67 and finally component 5 consists of with five components, upon closer investigation, we could
observations 68 to 71. It is interesting to note that, the observe that, there is an overlap of component 1 and 2 and
critical fault detection which is accurately predicted component 3 and 4. When these components are
similarly to other clustering algorithms as well.
reorganized we can observe much similar pattern to the
IV. RESULTS previous cluster analysis.
In this research, initially, we were hypothesized that two V. CONCLUSION
states in data. One is healthy data set, and the other is
This research started out as a test bed to benchmark
unhealthy data set. Using PCA and T2 statistic, we were able different machine learning algorithms for early fault
to fit our hypothesis states and able to detect the faults 31 detection using unsupervised learning. In our results, T2
observations ahead. Whereas, without a tool and just based statistic provided more accurate results compared to GMM
on data plots we could observe the trends only 11 method, and no hypothesis was required to identify the
observations ahead. As we moved on to fitting different
relationship between cluster and state. One of the main
unsupervised clustering algorithms, we found most of the benefits of this method is that, even when this is deployed to
clustering algorithms provided much more than the T 2 the manufacturing environment, with minimum or no
statistic.
domain knowledge, one can identify fault or critical
condition when compared to clustering analysis. On the
Using elbow method and nbClust package, we were able other hand in clustering, some information about the data is
to identify that the most optimal number of clusters that needed to name the clusters as healthy, warning or critical.
could be formed was three. Based on these results, when Clustering methodology is undoubtedly a better tool in
data was fitted in hierarchical clustering, K-means, and C-
detecting different levels of faults where T2 statistic would
means, the results were nearly identical. Based on the be challenging after certain levels. To emphasize this, when
previous knowledge of the data, we were able to identify the cost machine maintenance is expensive, clustering
would be a flexible option where machine health can be Supply Chains. Package.” T echnol. Sci., 29: 25–43. doi:
monitored continuously until a critical level is reached. 10.1002/pts.2185.
[3] Jay Lee, Hung-An Kao, Shanhu Yang, (2014) “ Service
Innovation and Smart Analytics for Industry 4.0 and Big Data
In conclusion of this study, although most algorithms
Environment ”, Procedia CIRP Volume 16, 2014, Pages 3-8
provided nearly similar results, each algorithm provided [4] Lebold M, T hurston M. “Open standards for condition -based
deeper insight into the data. Hence, if the application is just maintenance and Prognostic systems”. In: Proceedings of
to detect the faults, T2 statistic would be an excellent tool. MARCON 2001—fifth annual maintenance and reliability
But if fault detection needs to be performed under different conference, Gatlinburg, USA, 2001.
levels then, clustering algorithms would be a better choice. [5] Garcia E, Guyennet H, Lapayre J-C, Zerhouni N. “A new
industrial cooperative tele-maintenance platform”. Comput
Table 3: Summary results of all Models Ind Eng 2004;46(4): 851–64.
Heirarchi Model- Heirarchi Model- [6] Groba. C, Cech. S, Rosenthal. F., Gossling. A, “Architecture
Obs Actual T2 K-Means C-Means Obs Actual T2 K-Means C-Means
cal Based cal Based of the predictive maintenance framework”, 6th International
1 H 0 1 1 1 1 37 H 0 1 1 1 1
Conference on Computer Information Systems and Industrial
2 H 0 1 1 1 1 38 H 0 1 1 1 1
3 H 0 1 1 1 1 39 H 0 1 1 1 1
Management Applications, 2007, IEEE
4 H 0 1 1 1 1 40 H 0 1 1 1 1 [7] Ethem Alpaydin, “Introduction,” in Introduction to Machine
5 H 0 1 1 1 1 41 F 1 2 2 2 3 Learning,3rd ed. Cambridge
6 H 1 1 1 1 1 42 F 1 2 2 2 3
[8] Ahmed, M., Baqqar, M., Gu, F., Ball, A.D., 2012. “Fault
7 H 0 1 1 1 1 43 F 1 2 2 2 3
8 H 0 1 1 1 1 44 F 1 2 2 2 3 detection and diagnosis using principal component analysis of
9 H 0 1 1 1 1 45 F 1 2 2 2 3 vibration data from a reciprocating compressor”, in:
10 H 1 1 1 1 1 46 F 1 2 2 2 3 Proceedings of the UKACC International Conference on
11 H 0 1 1 1 1 47 F 1 2 2 2 3
12 H 0 1 1 1 1 48 F 1 2 2 2 3
Control, 3-5 September 2012, IEEE Press.
13 H 0 1 1 1 1 49 F 1 2 2 2 3 [9] Azzeddine Bakdi, Abdelmalek Kouadri, Abderazak Bensmail,
14 H 0 1 1 1 1 50 F 1 2 2 2 3 “Fault detection and diagnosis in a cement rotary kiln using
15 H 0 1 1 1 1 51 F 1 2 2 2 3 PCA with EWMA-based adaptive threshold monitoring
16 H 0 1 1 1 1 52 F 1 2 2 2 3
17 H 0 1 1 1 1 53 F 1 2 2 2 3
scheme”, Control Engineering Practice, Volume 66,
18 H 0 1 1 1 1 54 F 1 2 2 2 3 September 2017, Pages 64-75
19 H 1 1 1 1 1 55 F 1 2 2 2 3 [10] C. T . Yiakopoulos, K. C. Gryllias. I. A. Antoniadis, “Rolling
20 H 0 1 1 1 1 56 F 1 2 2 2 3 element bearing fault detection in industrial environments
21 H 0 1 1 1 1 57 F 1 2 2 2 3
22 H 0 1 1 1 1 58 F 1 2 2 2 3
based on a K-means clustering approach”, Expert Systems
23 H 0 1 1 1 1 59 F 1 2 2 2 3 with Applications Volume 38, Issue 3, March 2011, Pages
24 H 0 1 1 1 1 60 F 1 2 2 2 3 2888-2911
25 H 0 1 1 1 1 61 F 1 2 2 2 3
[11] Djurdjanovic D, Lee J, Ni J. “Watchdog agent, an infotronics-
26 H 0 1 1 1 1 62 F 1 2 2 2 3
27 H 0 1 1 1 1 63 F 1 2 2 2 3 based prognostics approach for product performance
28 H 0 1 1 1 1 64 F 1 2 2 2 3 degradation assessment and prediction”. Advance
29 H 0 1 1 1 1 65 F 1 2 2 2 4 Engineering Informatics 2003; 17(3–4): 109–25.
30
31
H
H
0
0
1
1
1
1
1
1
1
1
66
67
F
F
1
1
2
2
2
2
2
2
4
4
[12] Hotelling, H. (1933). “ Analysis of a complex of statistical
32 H 0 1 1 1 1 68 F 1 2 2 2 4 variables into principal components.” Journal of Educational
33 H 0 1 1 1 1 69 F 1 3 3 3 5 Psychology, 24, 417–441.
34 H 0 1 1 1 1 70 F 1 3 3 3 5 [13] IBM Israel, “Israel Electric corporation moves towards
35 H 0 1 1 1 1 71 F 1 3 3 3 5
36 H 0 1 1 1 1 72 F 1 3 3 3 5
smarter maintenance”, 2013, retrieved from www.IBM.com
[14] J. C. sridek, “Pattern Recognition with Fuzzy Objective
Function Algorithms”, New York: Plenum Press, 1981.
[15] Jacob Goldberger, Sam Roweis, “Hierarchical Clustering of a
VI. FUTURE SCOPE OF W ORK Mixture Model”, Neural Information Processing Systems
Fault detection is one of the preliminary analytics for Conference
predictive maintenance. Hence, detecting the fault [16] James R. Evans, Carl H. Lindner (2012), “ Business Analytics:
accurately is regarded important. This work is currently T he Next Frontier for Decision Sciences,” College of
performed for vibration data. The scope of this research can Business, University of Cincinnati, Decision Science Institute
[17] Jingli Yang, Yinsheng Chen, Zhen Sun, “A real-time fault
be extended out to other physics-based parameters and
detection and isolation st rategy for gas sensor arrays”,
combination of these parameters. It would also be Instrumentation and Measurement T echnology Conference
interesting to observe the detection accuracy for bigger (I2MT C), 2017 IEEE International, 22 -25 May 2017,
sample size and multiple fault states. 10.1109/I2MT C.2017.7969906
[18] Jolliffe, “ I.T . Principal Component Analysis”, Springer, New
York, 2002.
REFERENCES [19] P. Liggan and D. Lyons, “Applying Predictive maintenance
[1] Mobley, R Keith, “ An Introduction to predictive techniques to Utility Systems, Retrieved from Pharmaceutical
maintenance”, 2002, 2nd ed, ISBN 0-7506-7531-4 Engineering”, Official Magazine of ISPE, Nov/Dec 2011, Vol
[2] Battini, D., Calzavara, M., Persona, A., and Sgarbossa, F. 31 No.6
(2016) “ Sustainable Packaging Development for Fresh Food
 [20] Ramesh Sridharan, “Gaussian mixture models and the EM IEEE T ransactions on Automation Science and Engineering (
algorithm”, retrieved from Volume: PP, Issue: 99 )
https://people.csail.mit.edu/rameshvs/content/gmm-em.pdf [28] Zhimin Du, .Bo Fan..Xinqiao Jin, Jinlei Chi, “Fault detection
[21] Samaranayake, P. & Kiridena, S. (2012). “Aircraft and diagnosis for buildings and HVAC systems using
maintenance planning and scheduling: An integrated combined neural networks and subtractive clustering analysis
framework”. Journal of Quality in Maintenance Engineering, ”, Building and Environment Volume 73, March 2014, Pages
18 (4), 432-453. 1-11
[22] Stephen P. Borgatti, “How to Explain Hierarchical [29] Lindsay I Smith, “A tutorial on Principal Components
Clustering”, 1994, Connections 17(2):78-80 Copyright 1994 Analysis”, February 26, 2002, page 2-8.
INSNA [30] Malika Charrad, Nadia Ghazzali, Veronique Boiteau, Azam
[23] T . R¨ognvaldsson, S. Byttner, R. Prytz, S. Nowaczyk and M. Niknafs (2014). ”NbClust: An R Package for Determining the
Svensson, “Wisdom of Crowds for Self-organized Intelligent Relevant Number of Clusters in a Data Set. Journal of
Monitoring of Vehicle Fleets” 2014, IEEE Statistical Software”, 61(6), 1-36. URL
[24] T hamara Villegas, María Jesús Fuente, Miguel http://www.jstatsoft.org/v61/i06/
Rodríguez,”Principal Component Analysis for Fault Detection [31] Chris Fraley, Adrian E. Raftery, T . Brendan Murphy, and
and Diagnosis. Experience with a pilot plant”, Advances in Luca Scrucca (2012) “ mclust Version 4 for R: Normal
Computational Intelligence, Man-Machine Systems and Mixture Modeling for Model-Based Clustering,
Cybernetics, ISBN: 978-960-474-257-8
Classification, and Density Estimation T echnical Report No.
[25] Venkat Venkatasubramanian, Raghunathan Rengaswamy b,
KewenYinc, .Surya N. Kavurid, “A review of process fault 597”, Department of Statistics, University of Washington
detection and diagnosis: Part I: Quantitative model-based [32] Chris Fraley and Adrian E. Raftery (2002) “ Model-based
methods”, Computers & Chemical Engineering Volume 27, Clustering, Discriminant Analysis and Density Estimation ”
Issue 3, 15 March 2003, Pages 293-311 Journal of the American Statistical Association 97:611 -631
[26] Y. Yong, Z. Chongxun and L. Pan, “A Novel Fuzzy C-Means [33] Nagdev Amruthnath, T arun Gupta (2016), “Modified Rank
Clustering Algorithm for Image T hresholding”, Measurement Order Clustering Algorithm Approach by Including
Science Review, vol. 4, no.1, 2004. Manufacturing Data”, 4th IFAC International Conference on
[27] Yajun Wang, Fuming Sun, Bo Li, “Multiscale Neighborhood Intelligent Control and Automation Sciences, Reims, France,
Normalization-Based Multiple Dynamic PCA Monit oring June 1-3, 2016
Method for Batch Processes With Frequent Operations”,

View publication stats