See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/361415699

Building an algorithm for predictive maintenance

Article in UPB Scientific Bulletin, Series D: Mechanical Engineering · December 2020

CITATIONS READS

2 34

2 authors, including:

Mohammad Abiad
The American University of the Middle East
10 PUBLICATIONS 24 CITATIONS

SEE PROFILE

All content following this page was uploaded by Mohammad Abiad on 03 October 2022.

The user has requested enhancement of the downloaded file.

U.P.B. Sci. Bull., Series D, Vol. 82, Iss. 4, 2020 ISSN 1454-2358

BUILDING AN ALGORITHM FOR PREDICTIVE

MAINTENANCE

Mohammad ABIAD 1,2, Sorin IONESCU1

Predictive maintenance makes it possible to evaluate the operating condition

of equipment, helps in identifying failures, or predicts when the next possible error of
an equipment is going to happen. Whenever you can diagnose or expect equipment
error, you can schedule maintenance in ahead of time, effectively manage inventory,
minimize downtime, and enhance the operational potency.
This paper proposes an Algorithm and shows an application of
telecommunication equipment's Predictive Maintenance with the help of statistical
approaches that could be implemented for an effective maintenance plan, such as
Time To Failure and Equipment's Reliability.

Keywords: Predictive maintenance, Statistics, Time to Failure, Reliability.

1. Introduction

Total Productive Maintenance (TPM) is aimed to improve the performance

of equipment and its reliability, it is also considered as a key aspect in a quality
management system. It strives to increase productivity by investing in appropriate
maintenance to reduce losses such as Breakdowns, availability of equipment, minor
stoppages that affect equipment efficiency and reduced output quality. Equipment
maintenance plays an important role in Total Productive Maintenance, and there
are different types of maintenance that are involved such as “Reactive
Maintenance”, “Periodic Maintenance”, “Proactive Maintenance” and “Predictive
Maintenance”.
In this paper, an Algorithm for Predictive maintenance is proposed, and the
statistical methodologies for an effective maintenance plan wer used, such as Time
to Failure and Reliability calculations. In predictive maintenance, users and
manufacturers, are benefiting from studying the current state of the equipment, the
reason that causes failures to equipment and the forecasting the possible time of the
next failure. This will help in maintenance planning, since once you can predict a
failure, maintenance actions should be done trying to avoid having downtime which
will improve the operational efficiency and at the same time helps in organizing

1
Faculty of Entrepreneurship, Business Engineering and Management, University
POLITEHNICA of Bucharest, Bucharest, Romania, e-mail: sc.ionescu@gmail.com
2
College of Business Administration, American University of the Middle East, Kuwait, e-mail:
mohammad.abiad@aum.edu.kw
338 Mohammad Abiad, Sorin Ionescu

company’s stock. “Basic understanding of predictive maintenance is continuous

monitoring to avoid system breakdown, which will lead to maximize the time
interval between consecutive maintenance tasks and reduce the overall production
costs”, [1]. “Predictive maintenance is a set of strategies used by the company to
lower the impact damage of a component in a telecom site, which has an objective
of reducing the operational expenditure and enhance customer satisfaction” [2].
Designing a predictive maintenance plan needs a very well-designed
approach for prompt evaluation of the equipment's operating condition and
detection of emerging errors. Several factors must be taken into account such as the
sources of errors and the number of their occurrence of which we can get it from
sensors or even from the key elements of the machine. In addition, the physical
dynamics understanding of the system is needed which includes deep understanding
of the interactions between different machine signals, the working range of the
device and the nature of the readings. Finally, one should clearly think of the main
maintenance objective, like error retrieval or maintenance plan evolution. The
main purpose of this paper is to design predictive maintenance algorithms that can
be used to evaluate the condition of an equipment as well as setting-up an effective
maintenance plan withing an appropriate statistical analysis.
This papers is divided into five sections, Section 1: Introduction, Section 2:
The Proposed Algorithm Workflow and Algorithm Stages, Section 3: Statistical
Analysis used in the Proposed Algorithm, Section 4: Algorithm Application and
Simulation and Section 5: Conclusion.

2. The Proposed Algorithm Workflow and Algorithm Stages

A predictive maintenance plan is used to analyze the data provided from the
running equipment in order to detect any possible failure that can happen. The idea
of having an algorithm for predictive maintenance, is to have an effective plan that
lower the cost of maintenance and helps in determining the appropriate time for
handling maintenance tasks. A predictive maintenance system uses prediction and
tracking algorithms to make the algorithm's end findings available and workable to
end users performing the real maintenance duties.
Tracking algorithm is to check the current status of the equipment and
define the main factors for having the errors of the equipment. An algorithm for
tracking generates information metrics called indices. An index is any system
information characteristic whose function changes as the system deteriorates in a
predictable manner. Any amount obtained from the information that groups
comparable system status together and distinguishes distinct status may be a
condition index. Thus, by comparing fresh information against proven indices of
defective circumstances, a Tracking algorithm can conduct detection or diagnosis
of errors. This information are used to evaluate the equipment present condition and
 Building an algorithm for predictive maintenance 339

to recognize and help in identifying the errors. “Machine faults are measures of
imperfection that physically and systematically change as the machine transitions
towards failure” [3].
prediction algorithm is important to detect future possibility of having such
error and predict when such error might occur. This algorithm is based on
estimating the time of error that might occur, which depends on equipment's present
and previous status. Typically, a prediction algorithm forecast the Time-To-Failure
(TTF) of the equipment by evaluating the equipment's present status. Prognostic
method for machine degradation detection, can both assess machine performance
and predict the remaining useful life. Prediction may use modelling, machine
learning, or a mixture of both to forecast status indices' future values. These future
values will then be used to calculate TTF metrics, which will determine whether
and when maintenance should be carried out. The algorithm can then compare the
expected values with a limit that defines the equipment's good operation, predicting
whether and when the error occurs. A part of the prediction algorithm will be the
tracking algorithm, since prediction process will need data from current and past
records or sometime simulation is needed when the error is rarely to be happened.
A clear workflow should be stated at first for an algorithm so that its
implementation leads to a useful result in forming a predictive maintenance plan.

2.1. The Proposed Algorithm Workflow: A workflow to build a predictive

maintenance algorithm is shown in Figure 1.

Fig. 1. The Proposed Algorithm Workflow

In any algorithm, you will have to follow different stages which is

considered to be basics for running the algorithm. Hence, when you plan to create
a detection model (for tracking) or a forecast model (for prediction) starting with
data that defines your scheme in a range of good and defective circumstances will
be one of the major steps. Evolving such a model involves the identification of
suitable status indices and the preparation of a model for their interpretation. This
340 Mohammad Abiad, Sorin Ionescu

method is likely to be incremental as you attempt distinct indices and distinct

designs until you discover your application's best model. Lastly, the algorithm is
deployed and integrated into your equipment tracking and maintenance systems.
The Workflow preparation stage consist of five major points: The Data
Acquisition, The Data Cleaning, Defining the Indices, Building Models and
Algorithm Implementation.

2.2. Data Acquisition: “Data are the facts that are collected, analyzed and
summarized for presentation and interpretation. All the data collected in particular
study are referred to as the data set” [4]. Data is a set of information that is collected
from the experimental units of which could be identified as Humans, Machines,
Animal or anything under study. The first step in evolving predictive maintenance
algorithm is data collection. "The data acquisition process transforms the sensor
signals into domains that have the most information to represent the condition of
the equipment or a fusion of several domains" [5]. For a good plan of predictive
maintenance, data collection plays an important role of which the process of the
data collection might require a large data base in some circumstances when tracking
a time series data.
Data can be collected in three different status, Status 1: When the system is
normally working, Status 2: When the system is working with error conditions and
Status 3: When the system in not working. When regular maintenance for the
equipment is performed, then it will be very difficult to have errors and so we will
have limited data for failure datasets. For that reason, one could think of simulated
data. The more data you can collect, the more accurate (power) you will get and the
lower the level of significance will be.

2.3. Data Cleaning: "Data preparation is an indispensable step in order to

convert various data forms and types into proper format that is meaningful to
machine learning predictive model" [6]. After the data acquisition stage, an
important step in this algorithm is to transform your data into a form that makes it
easier for defining the indices. This will include cleaning data from outliers or even
from missing values of which this part is considered to be the simple technique.
Data cleaning is usually done by applying the “Data Explore” to summarize the
values of the data and then look for extreme values, which are usually defined as
values outside the expected range, then we investigate whether these values are
considered as outliers or not. As for the missing values, different methods can be
used, like we can replace the missing value by the computed average, in other cases,
it could be replaced by similar cases or even delete such cases, all depends on the
critical case of study and the historical data available for such cases.
Once you have a good knowledge about your equipment and the type of
data you have, the data cleaning technique will be easier to be selected. Statistical
 Building an algorithm for predictive maintenance 341

software such as SPSS can be used for the data cleaning issues by looking for
extreme values, replacing missing values or even reverse recoding when needed.

2.4. Define the Indices: Any characteristic that can be helpful to distinguish
between normal and improper operations or to predict Time to Failure can be an
index. The identification of indices in your system information, whose conduct
changes with the equipment degrading in a predictable manner, is one important
step in the growth of predictive maintenance algorithms. "Organizations face both
new opportunities and challenges, one of them is predictive analysis using computer
tools capable of detecting patterns in the analyzed data from the same rules that can
be used to formulate predictions" [7]. Examples of indices are some values resulting
from 'Simple analysis' by finding some numerical measures such as the average
over time, 'Complex analysis' that explains or describes the changes in field over
time, 'Model-Based analysis' when sometime looking for estimating parameters.
The most appropriate indices to be stated are related to your system
understanding as well as the sort of scheme that you have.

2.5. Building a Model: The tracking or prediction model is the core of the
predictive maintenance algorithm. This model examines derived indices to
determine the system's present condition (error tracking and identification) or its
prospective condition (Time to Failure forecast).

2.5.1. Error tracking and identification: The tracking and identification of

errors depends on the use of one or more conditions to determine whether the
function is safe or bad and between distinct kinds of errors. A straightforward fault-
tracking model is a limit value for the measure of a fault when it is surpassed. In
order to determine the probability of a specific failure situation, another model
could compare a condition index to a statistical distribution of measurement values.
One of the tracking measure that can be calculated is the Mean Time to
Repair (MTTR) measured by getting the total period needed for all repairs during
the year and then divided by the number of repairs made during the same period.
This measure will be also used in the predictive maintenance plan, since this period
should be taken into consideration when assigning a date for maintenance.
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = (1)
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅

2.5.2. Time to Failure Prediction: To predict the time to failure, a model

should be generated and as examples on such models we can see the following:
• A model that corresponds to a condition index's time progress and foresees how
long it will take before the condition index reaches certain limit values to
indicate an error.
342 Mohammad Abiad, Sorin Ionescu

• A model which matches up the time progress of a condition index from failing
systems to calculated or computer generated time series. Such a model can
calculate the present system's most probable time to fail.
The Mean Time between Failure (MTBF) is an estimate of the reliability of
the equipment. It is calculated by using the total operated time divided by the total
number of failures. In addition, one can calculate the failure rate of the equipment,
which is 1 divided by the MTBF to get an idea about the failure rate per year of this
equipment. Such measures are used for prediction model.
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 , (2)
In many cases, the Time to Failure might be predicted by using dynamic
system models, using many techniques such as threshold and survival analysis of
which the Weibull Distribution is mainly used.
“From a maintenance process perspective, Remaining Useful Lifetime
estimation is very useful for long-term planning of spare parts supply and
maintenance scheduling. On the other hand, failure prediction is more useful for
handling unexpected failures that might happen in a short time-span” [9].

2.6. Algorithm Implementation: Once you are done with your data by
processing it properly and then by generating an appropriate prediction model, you
use your algorithm and incorporate it into your system. In addition, according to
your systems’ technical and other details, you decide whether to implement your
algorithm on the cloud, storage arrays or a mix between them by using the storage
arrays for pre-processing and when setting the algorithm properties, whereas use
the cloud for running the predictive model.

3. Statistical Analysis used in the Proposed Algorithm

In the main heart of the proposed algorithm there are different probability
distribution could be used in order to prepare studying the equipment reliability and
plan for predictive maintenance. In this section, the most commonly used
probability distributions are summarized, followed by the technique used to
determine the best fit distribution and then summarizing the way how to calculate
the equipment reliability. Such Analysis are used to determine the key indices of
the equipment and getting an idea about the efficiency level of the equipment before
building a tracking model of even predicting the time to failure or what is also called
as Remaining Useful Lifetime.

3.1. Commonly used Probability Distribution: There are many

probability distributions that are used in industrial engineering, we are going to list
in this report the most commonly used probability distribution in the domain of
telecommunication sector.
 Building an algorithm for predictive maintenance 343

3.1.1. Exponential Distribution: “The exponential distribution is one of the

most commonly used distributions in reliability and is generally used to predict the
probability of survival to time t” [10]. Although the exponential distribution is
commonly used to model component repair, it is not well suited for this task. It is
relatively accurate representation of electronic component time to failure and
should only be used for reliability prediction during the constant failure rate or
random failure phase of operation. The exponential Distribution has the parameter
𝜇𝜇 = 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 and CDF (Cumulative Distribution Function):
1
− 𝑡𝑡
𝐹𝐹(𝑡𝑡) = 1 − 𝑒𝑒 𝜇𝜇 𝑓𝑓𝑓𝑓𝑓𝑓 𝑡𝑡 ≥ 0 (3)

Noting that the sum of individual exponential random variables is an

exponentially random variable.

3.1.2. Normal and Lognormal Distribution: “The normal (Gaussian) and

lognormal distributions are continuous statistical distributions that can be used to
model a large number of varying types of system repair behavior” [11]. In
telecommunications systems, the failure can many times be well represented by the
exponential distribution, whereas, Repair is more often well modeled by normal or
lognormal random variables. The lognormal distribution is simply the distribution
of a random variable whose logarithm is normally distributed. It would be very
unusual to use a normally distributed random variable to model the time to failure
of a component in a telecommunications system. Exceptions to this might occur in
submarine cable systems or wireless propagation models.
The Normal Distribution has the following parameters: 𝜇𝜇 (the mean) and 𝜎𝜎 (the
standard deviation); The C.D.F is given by:
𝑡𝑡−𝜇𝜇 𝑡𝑡−𝜇𝜇
𝐹𝐹(𝑡𝑡) = 𝑃𝑃 �𝑇𝑇 ≤ � = 𝜑𝜑 � � (4)
𝜎𝜎 𝜎𝜎

3.1.3. Weibull distribution: "Generally, all but the most sophisticated

telecommunications systems failure performance models use exponentially
distributed time to failure owing to the memory less property" [11]. The parameters
of the distribution provide a great deal of flexibility to model systems in which the
number of failures increases with time (bearing wear), decreases with time (some
semiconductors), or remain constant with time (failures caused by external shocks
to the system). The Weibull Distribution has the following parameters: 𝛿𝛿 =
𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 and 𝛽𝛽 = 𝑆𝑆ℎ𝑎𝑎𝑎𝑎𝑎𝑎 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 With CDF (Cumulative
Distribution Function) [12]:
𝑡𝑡 𝛽𝛽
�
𝑭𝑭(𝒕𝒕) = 1 − 𝑒𝑒 −�𝛿𝛿 𝑓𝑓𝑓𝑓𝑓𝑓 𝑡𝑡 ≥ 0 (5)
344 Mohammad Abiad, Sorin Ionescu

3.2. Technique used to determine the Best Fit Probability Distribution:

Data might follow different probability distributions; By using Statistical software
such as SPSS or MINITAB, the "Individual Distribution Identification" is able to
determine the best-fit distribution of the given data.
After running the “Individual Distribution Identification”, at first, one
should look for the significant p-value (𝑃𝑃𝑣𝑣 > 0.05) to check whether this
distribution fits the data, then for significant distribution we use the Anderson-
Darling parameter to check the Best Fit Distribution. The Lowest the Anderson-
Darling, the best the distribution will be.
As an example, Figure 2 shows the result of the best fit distribution when
studying the distribution of ‘Time to Repair a submarine Fiber optic Cable’
Probability Plot for TTR
Goodness of Fit Test
Normal - 95% CI Exponential - 95% CI
99.9 99.9
Normal
99 90 AD = 1 1 .046
50 P-Value < 0.005
90
Percent

Percent

10 Exponential
50
AD = 0.734
10 P-Value = 0.251
1

1 Weibull
0.1 0.1 AD = 0.500
0 60 120 180 0.01 0.1 1 10 100
P-Value = 0.220
TTR TTR
Lognormal
Weibull - 95% CI Lognormal - 95% CI AD = 1 .390
P-Value < 0.005
99.9 99.9
90 99

50
90
Percent

Percent

10 50

1 10

0.1 0.1
0.01 0.1 1 10 100 0.1 1 10 100 1000
TTR TTR

Fig. 2. Studying the best fit distribution for Time to Repair Data set

Result shows that the given data fits the Exponential and Weibull
Distribution (since both 𝑃𝑃𝑣𝑣 > 5%), so no evidence to reject the hypothesis of fitting
the distribution. Hence, it is better fit of Weibull distribution as it has the lowest
Anderson-Darling statistic (AD) value (AD=0.500).

3.3. Equipment’s Reliability: Analyzing equipments' reliability helps

organizations in general to improve and provide high quality service for their
customers. This will also help to gain an excellent relationship with their
customers.. Reliability is attained after achieving the desired outcomes required
from a designed system. Indeed, reliability is best defined as “the duration or
probability of failure-free performance under stated conditions” [11]. However, the
probability that an item will perform its intended function for a specified interval
understated environmental conditions also reflects a complementary definition for
the core meaning of reliability. In advance, the reliability function is a mathematical
expression, which analytically relates the probability of success to time, and is
defined by the following expression [13]:
 Building an algorithm for predictive maintenance 345

𝑅𝑅(𝑡𝑡) = 1 − 𝐹𝐹(𝑡𝑡) 𝑓𝑓𝑓𝑓𝑓𝑓 𝑡𝑡 > 0 (6)

Of which 𝐹𝐹(𝑡𝑡) is the cumulative probability function (CDF) of the Time to

Failure (TTF).

4. Algorithm Application and Simulation

As an Application of this predictive maintenance algorithm, we will talk

about the Remote Radio Unit (RRU) in the telecommunication sector.
An RRU is defined as “A wireless base station (also known as a cell site or
wireless base transceiver station, BTS) is a piece of equipment that facilitates
wireless communication between user equipment (UE) and a network” [14]. This
RRU can handle calls for a certain range that needs to be defined by the
manufacturer and experts as well. When the number of calls exceeds this range it
goes for warning level before the Threshold value, that is, before the Error occurs
which causes Drop Call.
The main purpose of our Algorithm is to check and predict the failure time of this
RRU equipment in order to plan for maintenance accordingly. The Input and Output
result of the proposed algorithm should contains at least the following Information:

4.1. Input Information:

1. Data Set: Historical data is needed which include the equipment’s overload
value that is monitored along with the equipment status of the equipment.
2. Equipment’s overload threshold value as well as a Warning level of such
overload, which is usually defined by the domain expert or the manufacturer.

4.2. Output Information:

1. A graph that shows the performance of the equipment in the historical data,
showing the threshold level at the same time and some warnings the equipment
might has when reaching a certain level of overload.
2. A warning message from the observed values will pop-up when reaching a
certain level of indicator.
3. A Message for detailed information about the equipment that helps the user to
determine the current status of the equipment along with the estimated date of
the next error so that a predictive maintenance is applied.

A random data set has been generated for this example and presented in table 1,
which include the historical data of day by day equipment’s overload records for 1
year along with the equipment, and check the equipment’s overload value based on
the defined regular load of the equipment that is defined by experts.
346 Mohammad Abiad, Sorin Ionescu

Input Information:
1. Data: Day by Day Equipment’s Status and Overload Value (Table 1).
Table 1: Day by Day Equipment’s Overload Value & Equipment Status
Day # Equipment’s Overload Value (in thousands) Equipment Status
1 7 Working
… … …
32 16 Warning
… … …
330 28 Error
… … …
365 4 Working
2. Equipment’s Overload Warning value = 15 (defined by an expert)
3. Equipment’s Overload Threshold value = 25 (defined by an expert)

Output Information:
1. Graph output (Figure 3): This graph summarize the current status and shows
the history of the equipment’s overload values, it highlights the bars as Red for
Errors, Green for the warnings and Blue for the normal status.

Fig. 3. Equipment’s Tracking Information for Equipment’s daily Overload

In this figure, we can see that there are 3 failure status in the monitored year
and there are more than overload value that falls within the warning range.
2. Warning Message (Table 2): this message will pop-up once an equipment’s
Overload value exceeds 15 but still below the threshold value and gives you
the following summary:
Table 2: Warning Message
Warning Date 18-Dec-18
Equipment’s Overload Value 17
Last Warning Date 6-Dec-18
Number of Warnings since Last Error 2
3. Detailed Information Message (Table 3): This will include the equipment
current reliability, the estimated time to failure which is computed based on the
appropriate probability distribution of the Equipment’s Overload variable.
 Building an algorithm for predictive maintenance 347

Table 3: Detailed Information Message

MTBF 110.67 days
Date of Requested Information 31-Dec-2018
Estimated Date for the Next Error 16-Mar-19
Number of Days Left before the
75 days
Expected Error
Equipment Reliability 86.9%
The Pop-up Message with such information, are calculated based on the input data,
which applies the appropriate calculation using the best fit probability distribution
of the data. This Pop-up message will give some details about the Mean Time
Between Failure (MTBF) based on the historical data of errors occurred, as well as
the expected date for the next error along with the equipment reliability which is
the % that the equipment will be reliable before the next expected error.

5. Conclusions

Evaluating the current condition of an equipment, as well as predicting the

possible time of having a failure in the future, are the main targets of companies in
order to improve their efficiency and decrease the downtime possibilities they might
face. Predictive maintenance if applied properly would help reaching such goals
with the application of well-designed algorithm for that issue.
This paper suggested workflow for the algorithm evolution with an
explaination of each suggested stage, the Data Acquisition, Data Cleaning, Forming
the Indices, Building Models and Algorithm Implementation. The use of statistics
is very important when building the tracking and prediction algorithm. For this
reason, a brief of the most commonly used probability distribution in this domain
has been summarized along with their cumulative probability function which is
used in the calculation of equipment's reliability.
Finally, an algorithm application and simulation has been stated to
summarize the input and output information of the algorithm, the warning and
detailed message that can be generated about the equipment which includes
information about the mean time between failures, as well as the next expected error
along with the equipment reliability. This will help the use to build an effective plan
for the maintenance of which we call it the predictive maintenance plan.

REFERENCES
[1]. Motaghare, O., Pillai, A.S. and Ramachandran, K.I., 2018, December. Predictive Maintenance
Architecture. In 2018 IEEE International Conference on Computational Intelligence and
Computing Research (ICCIC) (pp. 1-4). IEEE.
 348 Mohammad Abiad, Sorin Ionescu

[2]. Abiad, M., Kadry, S. and Ionescu, S., 2018, September. Preventive & Predictive Maintenance
of Telecommunication Equipment-A Review. In 2018 4th International Conference on
Applied and Theoretical Computing and Communication Technology (iCATccT) (pp. 154-
159). IEEE.
[3]. Swanson, D.C., 2001, March. A general prognostic tracking algorithm for predictive
maintenance. In 2001 IEEE Aerospace Conference Proceedings (Cat. No. 01TH8542) (Vol. 6,
pp. 2971-2977). IEEE.
[4]. Anderson, D. et al: Statistics for business & economics 4th ed, SOUTH-WESTERN CENGAGE
Learning, 978-1-4737-2656-7, Andover, (2017).
[5]. Li, Z., Wang, Y. and Wang, K.S., 2017. Intelligent predictive maintenance for fault diagnosis
and prognosis in machine centers: Industry 4.0 scenario. Advances in Manufacturing, 5(4),
pp.377-387.
[6]. Munirathinam, S. and Ramadoss, B., 2016. Predictive models for equipment fault detection in
the semiconductor manufacturing process. IACSIT International Journal of Engineering and
Technology, 8(4), pp.273-285.
[7]. Candanedo, I.S., Nieves, E.H., González, S.R., Martín, M.T.S. and Briones, A.G., 2018, August.
Machine learning predictive model for industry 4.0. In International Conference on Knowledge
Management in Organizations (pp. 501-510). Springer, Cham.
[8]. Li, Z., Zhou, S., Choubey, S. and Sievenpiper, C., 2007. Failure event prediction using the Cox
proportional hazard model driven by frequent failure signatures. IIE transactions, 39(3),
pp.303-315.
[9]. Aggarwal, K., Atan, O., Farahat, A.K., Zhang, C., Ristovski, K. and Gupta, C., 2018, December.
Two Birds with One Network: Unifying Failure Event Prediction and Time-to-failure
Modeling. In 2018 IEEE International Conference on Big Data (Big Data) (pp. 1308-1317).
IEEE.
[10]. Durivage, M.: Practical engineering, process, and reliability statistics, ASQ The Global Voice
of Quality, 978-0-87389-889-8, Milwaukee, (2015).
[11]. Ayers, M.: Telecommunications System Reliability Engineering, Theory and Practice, John
Wiley & Sons, 978-1-118-13051-3, Hoboken, (2013).
[12]. Montgomery, D.; Runger, G.: Applied statistics and probability for engineers 6th ed, John
Wiley & Sons, 978-1-118-74412-3, Singapore, (2014).
[13]. Hafaifa, A. et al: Reliability analysis using Weibull distribution applied to a booster pump used
in oil drilling installations. Journal of the Technical University Sofia, Branch Plovdiv,
Bulgaria, Fundamental Sciences and Applications, Vol. 22, (2016), pp.31-37, ISSN 1310 –
8271.
[14]. Wireless Base Station - Remote Radio Unit (RRU) | IDT. (2019). Retrieved 14 October 2019,
from https://www.idt.com/us/en/application/wireless-infrastructure/wireless-base-station-
remote-radio-unit-rru

View publication stats