Syntax Literate:
Jurnal Ilmiah Indonesia p–ISSN: 2541-0849 e-ISSN: 2548-1398
Vol.
9, No. 3,
Maret 2024
PREDICTIVE MAINTENANCE USING LINEAR REGRESSION
Benedict Ariel
Kurnianto,
Rudy Hartono Prayogo*, Nidia
Pialina Nababan, Suharjito
Industrial Engineering Department, BINUS Graduate
Program – Master of Industrial Engineering, Bina Nusantara
University, Indonesia
Email: [email protected]*
Abstract
Problems regarding machine damage often
occur in many industries, especially the manufacturing industry, which causes
large losses for companies. This is of course influenced by various factors
such as engine
temperatures that are too high, engine
rotation that is too fast, poor engine
torque values, and so on. This research
aims to provide predictive analysis
results regarding engine
conditions that have the potential to experience damage. To achieve this goal,
this research will carry out predictive maintenance
analysis using a linear
regression analysis approach in which two
linear regression models will be carried out where the first model
involves PCA preprocessing and the second model is carried out without PCA.
This research will use the predictive maintenance dataset
from the conference (Matzka, 2020). It is known that the MSE, RMSE, MAE, and R2 values of the two methods have the same values, namely 0.909, 0.953, 0.806, and 0.772 respectively. Based on this research, it is concluded
that whether PCA is performed or not, it does not significantly affect
the results of the regression analysis. This outcome
can be attributed
to the artificial nature of the dataset, rendering it ideal. Moreover, the retained PCA value of 98% is close to the number of
attributes in the original dataset.
Keywords: Data Science, Predictive Maintenance, Manufacturing Industry, Linear Regression, Principal Component Analysis, Orange Data Mining Software
Introduction
The advancement of technology allows us to easily acquire large
volumes of data, often referred to as big data. This is undoubtedly positive
news for companies, as they can leverage this data to enhance their
performance. To harness the power of this data, a
more in-depth analysis
is necessary, and this is where data science comes into play. According to (Aalst, 2016), data science is an interdisciplinary field with the goal
of turning data into tangible value. Data science simplifies the analysis of big data, ensuring accurate
results in line with the
acquired data (Aalst, 2016).
Data science
analysis is commonly
applied across various sectors to examine the data at hand. One crucial sector
where data science plays a significant role is manufacturing,
particularly in the context of predictive maintenance. The manufacturing
industry often faces issues stemming from various factors. These issues are
critical to address because equipment breakdowns can result in reduced
production efficiency, ultimately leading to unfulfilled customer demands
(Pranoto, Matondang,
& Siregar, 2013). This is why predictive
maintenance is of great importance, as it helps minimize equipment failures,
thus preventing production stoppages (Pranoto, Matondang, & Siregar, 2013).
In
response to these challenges, this research focuses on conducting predictive maintenance analysis using
regression analysis methods. Additionally, two approaches will be
implemented: one involving Principal Component Analysis (PCA) as an initial
step and the other without Principal Component
Analysis (PCA). This research is significant, as it
provides an opportunity for the academic community to contribute new knowledge
across various disciplines. It serves as a valuable source of insight
into the application of cutting- edge technologies like machine learning
and data processing algorithms in data science.
Furthermore, it offers the broader public a new technique for
addressing their machine maintenance issues, thus reducing potential losses.
Many
studies on maintenance-related issues have been conducted in recent decades (Selcuk, 2017). According to Heizer and Render, the definition of maintenance includes
all activities involved in keeping the equipment system in line with the work order,
with maintenance activities divided into two
types: preventive maintenance and corrective maintenance. Corrective
maintenance, as defined by Hendarsin, is maintenance
performed only when a component/system fails. On the other hand, according to
(Ebeling, 2010), preventive maintenance is scheduled maintenance activities,
generally performed periodically, involving activities like inspection, repair,
replacement, cleaning, lubrication, adjustment, and alignment.
In
addition to these two types of
maintenance, predictive maintenance, or prediction-based maintenance, is
receiving increasing attention in research related to data acquisition,
infrastructure, storage, distribution, security, and intelligence (Zonta,
et al., 2020). Predictive maintenance, as known, is a maintenance strategy that leverages technology, data analysis,
and an understanding of equipment performance to predict when equipment
is likely to fail or experience issues.
Its goal is to identify potential issues or failures before
they occur so that corrective or maintenance actions
can be taken promptly (Senanayaka, et al., 2022). Predictive maintenance is
considered to be a more efficient strategy compared to other maintenance
approaches like preventive or corrective maintenance because it is only
performed when necessary. Over time, research in predictive maintenance has
advanced due to its ability to generate predictions based on equipment performance or condition, which is crucial for the future of the industry
(Wu, Jennings, Terpenny, & Kumara,
2016).
The key requirement for effective predictive maintenance is to have an adequate amount of data from every
part of the manufacturing process. Properly conducted analysis can reduce
maintenance costs, lead times, and improve productivity and quality (Kiangala & Wang, 2018).
Regression
analysis and predictive maintenance are two complementary concepts in the world
of industry and maintenance management. A statistical technique called
regression analysis is used to determine how one or more independent variables,
or predictors, relate to a dependent variable.
In the context of predictive maintenance, regression
analysis can assist in predicting the remaining lifespan of equipment based on
certain factors (Montgomery, Peck, & Vining, 2021). On the other hand,
predictive maintenance is an approach that enables organizations to predict
when equipment is likely to fail, allowing maintenance to be performed before
failure occurs (Mobley, 2002).
Regression
analysis is a statistical technique used to understand the relationship between
independent variables (predictors) and a dependent variable (response). It
allows measuring to what extent changes in one
variable will affect another variable. Regression is a widely used
tool in various disciplines, including social sciences, economics, natural
sciences, and business. One crucial concept in regression analysis is the
regression coefficient. This coefficient measures how much the independent
variable influences the dependent variable. In a simple
linear regression equation
(Y
= a + bX + ε), "b" is the regression coefficient. This coefficient describes how changes in the independent
variable (X) will affect changes in the dependent variable (Y). But in order to
apply regression analysis effectively, certain presumptions must be met: the
independent and dependent variables must have a linear relationship; changes in the independent variable
must affect the dependent variable proportionately;
homoscedasticity—that is, the variance of the random errors must remain constant along the
regression line—must be assumed; and, lastly,
the residuals, or errors, are supposed to have a normal distribution.
In
the context of predictive maintenance, regression analysis can play a vital role. Regression analysis
can be used to model the relationship between
monitoring variables (such as
vibration or temperature) and equipment failure
trends. By analyzing
historical data, regression can be used to understand
how monitoring variables
correlate with equipment condition changes. For example, regression analysis can be used
to identify the monitoring variables most strongly correlated with equipment
failures. This can help build a stronger predictive model for forecasting the
remaining lifespan of equipment. By knowing
the relationship between
monitoring variables and failures, early signs of damage can be identified,
and preventive maintenance can be performed before more severe damage occurs. Moreover,
regression analysis
can also be used in evaluating the impact of maintenance that has been conducted. For instance, regression
analysis can be used to understand whether the maintenance performed has
resulted in a significant improvement in equipment performance. This can help assess the effectiveness of predictive
maintenance programs.
Principal
Component Analysis (PCA)
A statistical technique
called principal component analysis
(PCA) uses advanced
mathematical concepts to break down a
group of possibly correlated variables into a smaller set of variables called
principal components. The goal of PCA is to
handle high-dimensional data by
reducing the complexity of the data (Richardson, 2009). There are several advantages to working with a smaller set of data rather than the original
high-dimensional data. According to (Kherif & Latypova,
2020), these advantages include:
1.
The ability
to visualize data in 2D or 3D
2.
Reduced storage space
3.
Elimination of collinearity
4.
Reduction of noise
New
components are produced by PCA analysis. The projection of the original data onto the first main axis provides
the first principal component, which captures
the majority of the variation in the data. The majority
of the variation in the data that the first
principal component was unable to explain is explained by the second principal
component, which is the
projection of the original data onto the second principal axis. Until the whole data matrix
is deconstructed, each succeeding
principle component explains the majority of the variance under the restriction
that it is orthogonal to the preceding principal components.
Creating
disease models that satisfy these requirements can be greatly aided by PCA.
Furthermore, PCA has the distinct benefit of offering a condensed representation of data that captures
the essential elements
of individual variations in the age of Big Data and growing interest in customized treatment.
This
research will also conduct a review of several previous studies where 8 studies
were obtained which are shown in Table 1 along with the author, title, method and main
results.
Table 1. Review
of previous research
Author |
Method |
Results |
(Chazhoor, et.al, 2020) |
Using various classification models. |
Logistic Regression is the suitable
model for the dataset. |
(Susto, et.al, 2015) |
Employing multiple classifiers such as SVM
and KNN for classification. |
Multiple Classifier PdM- SVM exhibits
better performance compared to other approaches. |
(Bukhsh, et.al, 2019) |
Using Decision Tree, Random Forest, and Gradient Boosted Tree. |
GBT Model is the most optimal results with an accuracy of 86% for maintenance needs prediction. |
(Abidi, Umer, et.al, 2020) |
Applying a variety
of machine learning models such as NN, SVM,
KNN, and RNN. |
Proposed Method SH-WOA
is the best model. |
(Hadi, Hady, et.al, 2023) |
Using Random Forest, Gradient-Boosting Classifier, Extra Trees, Light
Gradient-Boosting Machine, and Extreme
Gradient Boosting. |
Random Forest achieves high accuracy
with 99.7% on the test set. |
(Senanayak a, et al., 2022) |
Using ANN, SVM, TLNN,
and SiMuS-TL. |
SiMuS-TL attains maximum classification accuracy above an unknown target, supporting the effectiveness of the proposed method. |
(Selvaraj, et.al, 2022) |
Using KNN, XGBoost, ITD-SVM, GAN-SAE, and a Proposed Model. |
The proposed model can produce
good results without
the need for
time-consuming feature
engineering and complex signal-transmitting tasks. |
(Kim, Park, & Jung, 2021) |
Comparing SVM and Random Forest with CS_SVM and CS_RF. |
One-Class Method proves more effective at detecting errors when faulty data is scarce compared to binary methods. |
In previous
research such as that conducted
by (Chazhoor,
Y, M, Sanjana, & R, 2020) it was discovered that the algorithms used to
model classification in carrying out predictive maintenance were Random Forest,
Logistic Regression, Decision Tree, and Multi-layered Perception. Apart from
that, research conducted by (Susto, Schirru, Pampuri, McLoone, & Beghi, 2015) only used two algorithms to analyze classification,
namely SVM (Support Vector Machine) and KNN (K-Nearest Neighbors). It is known
that most studies directly carry out classification analysis by comparing
several different algorithms.
From several
studies, it is known that most of the research was carried out using classification methods.
Apart from that, there are not many studies that carry out the PCA process first before making predictions. Based on
this, two research gaps were formed that differentiate this research from
previous research: (1) this research will carry out regression analysis which aims to
increase maintenance efficiency, reduce unplanned machine downtime, and optimize
the use of maintenance resources, and (2) this research will carry
out a comparison where the methods being compared
are using the PCA method first and then carrying out regression analysis and
the second method is without using PCA and directly carrying out regression
analysis. Moreover, this research aims to provide predictive analysis results
regarding engine conditions that have the potential to experience damage.
Research
Methods
Research
will be carried out by formulating the problem and background regarding predictive maintenance. After that, we will determine the objectives of
the research that will be carried out so that the research has the right
objectives and answers the problems that are currently occurring. In order for research
to be reliable, it is necessary to study previous research, after which gaps will be determined that can be used
as
a differentiator between this research and previous research.
Data collection will be carried out using a dataset regarding predictive
maintenance originating from the conference (Matzka,
2020). After that, the analysis will be carried out twice, where the first
analysis is carrying out the PCA stage first
and after that the linear
regression analysis is carried out. Then the second analysis is without
doing PCA, but directly doing linear regression analysis. The analysis
will be carried out in the Orange application. After
the analysis is carried out, conclusions can be drawn and determine the appropriate linear regression for the dataset
that has been obtained. The research
framework can be seen in Figure 1.
Figure 1. Research
flow
In this
research, a dataset regarding predictive maintenance taken from a conference (Matzka, 2020) will be used. From this dataset, it is known
that there are 10,000 instances with a total of 14 attributes consisting of:
1. UID: unique identifier that ranges from 1 to 10000
2. Product
ID: consists of the letters
L, M, or H for low (50% of all products), medium (30%), and
high (20%) product quality as a product quality variant and a specific serial
number for that variant
3. Type: only product
types L, M, or H from column 2
4. Air Temperature [K]: produced by a random
walk procedure, standardized at around 300 K to a standard deviation of roughly
2 K.
5. Process Temperature [K]: produced by adding 10 K to the
air temperature plus a random walk procedure standardized to a 1 K standard
deviation.
6. Rotational Speed [rpm]: computed by adding
normally distributed noise to 2860 W of electricity.
7. Torque [Nm]: There are no negative values
and the torque value is generally
dispersed around 40 Nm with an SD of
10 Nm.
8. Tool Wear [min]: The process tools used in H/M/L quality versions have an additional 5/3/2 minutes
of wear.
9. A value of 1 in the "Machine
Failure" column denotes a machine failure, whereas
a value of 0 denotes
none at all.
10. Tool Wear Failure (TWF): 120 times in our
data set, the tool will either fail or be replaced at a randomly chosen tool wear time of 200–240 minutes.
The tool was changed
69 times during this period,
and it failed 51 times (chosen
at random).
11. Heat Dissipation Failure
(HDF): If the tool rotation
speed is less than 1380 rpm and the difference between the process and
air temperatures is less than 8.6 K, heat dissipation will lead to process
failure. This happens in 115 different data points.
12. Power Failure (PWF): the outcome of torque
and rotational speed (measured in rad/s) that match the process's power requirements. The process will fail if this
power is less than 3500 W or more than 9000 W, as observed in 95 instances in
our data set.
13. Overstrain
Failure (OSF): The process will fail as a result of excess stress if the product
of torque and tool wear
for product variant L (12,000 M, 13,000 H) exceeds 11,000 minNm.
98 data points are covered by this.
14. Random Failures (RNF): Regardless of the
process parameters, there is a 0.1% chance that any process will fail. Less
frequently than one might anticipate for the 10,000 data points in our data set, this only happens
on 5 of them.
Descriptive Analysis
The descriptive results of the analysis of each attribute
are given, where the results of the distribution, mean, mode, median, dispersion, minimum, maximum and missing values of each attribute are known. The
results of the descriptive analysis can be seen in Figure 2.
Figure 2. Descriptive analysis
of dataset
The initial stage of data processing is using the correlations widget to determine the relationship between variables, after which imputation will be carried out
on the dataset to determine whether there are missing
values or not, which will be analyzed for the presence or absence of outliers so that they will be removed using the concatenate widget. After that, two
analyzes will be carried out, where first the PCA process will be
carried out, after which a linear regression analysis will be carried out. Then the second analysis
does not go through the PCA process, but directly carries out
linear regression analysis. The two analyzes will be compared to find out the
best results. Orange software will be used to perform linear regression analysis. The working framework for
Orange can be seen in Figure 3.
Figure 3. Linear regression used for predictive maintenance
The
first step in carrying out the analysis is to enter the Predictive Maintenance
dataset obtained from the Kaggle website
from the conference
(Matzka, 2020). In this dataset, it is
known that there are 10,000 instances with a total of 14 attributes. Detailed
information about the dataset can be seen in Figure 4.
Figure 4. Dataset
of predictive maintenance
Before
conducting the analysis, data transformation was performed on the attributes
Machine Failure which were originally categorical types, converted into numeric to enable
a proper linear regression analysis.
Figure 5 displays
the data transformation's
outcomes.
Figure 5.
Transformation Data
Next, start selecting the attributes that will be used. In this
case, attributes that have a subset of numerical categories will be selected for the next analysis process.
This process can be
seen in Figure 6.
Figure 6. Selecting numeric
attribute
After selecting the attributes you want to use, it is important to first know
the relationships that occur between variables to make the analysis process easier. The relationship
formed can be seen in Figure 7.
Figure 7. Correlations analysis
Impute
At this stage, an analysis
of the missing values in the dataset will be carried out, where the missing values
will be replaced with the average number of the column. The settings
can be seen in Figure 8.
Figure 8. Impute Analysis
Descriptive Analysis of Numeric Attribute
In
this case, an analysis will be carried out on each previously selected
attribute to determine the distribution, mean, mode, median, dispersion,
minimum, maximum and missing values of each attribute. These results can be seen in Figure 9.
Figure
9. Descriptive analysis
of numeric attribute
is
The next step is to analyze
outliers on the attributes
selected in the previous step. At this stage, the method used is
Local Outlier Factor with a contamination parameter
of 10% with neighbors
20 and Euclidean metrics. After processing the outliers,
it was discovered that the number of instances was reduced to 9103 data. In Figure 10, the outlier process displayed.
Figure 10.
Outlier’s analysis
Next, the analysis will be carried
out in two ways, namely through the PCA process and not
through the PCA process, where the results will be compared to determine the
best result.
1. Linear Regression Analysis with PCA
At this stage the process will begin by carrying out a PCA analysis where the 5 attributes will be reduced to 4 attributes or principle components with the names PC1, PC2, PC3, and PC4 which maintain 98% of the data
variation so that it can meet the requirements. Then the cumulative variance formed is
0.980 and the component variance is 0.184. The PCA process can be seen in
Figure 11.
Figure 11. Principal component
analysis
After that, attribute selection is carried out using the select
column widget where Air Temperature will be used as the target or dependent variable
(Y), while PC1 (X1), PC2 (X2), PC3 (X3),
and PC4 (X4) will be used as features or independent variables (X). Details of attribute selection can be seen
in Figure 12.
Figure 12. Selecting main components of PCA
After
attribute selection is carried out, linear regression analysis can be carried
out. The results of linear regression can be
seen in Figure 13.
Figure. 13. Result of linear regression with PCA
From these
results it is known that the linear regression equation formed is:
𝑌 = 300.003 + 0.033𝑋1 + 0.874𝑋2 − 1.435𝑋3 − 0.482𝑋4
From
this equation it is known that Air Temperature, which is the dependent
variable, is influenced by 4 independent variables, namely
PC1, PC2, PC3, and PC4. PC1
has a coefficient of 0.033. There is a positive correlation between PC1 and air temperature, as indicated by this positive
coefficient. It can be concluded
that if the PC1 value increases,
the Air Temperature value will increase too. The coefficient for PC2 is 0.874.
There is a positive correlation between PC2 and air temperature, as indicated by this positive coefficient.
It
can be concluded that if the PC2 value increases, the Air Temperature value will increase
too. PC3 has a coefficient of -1.435. This
negative coefficient indicates that there is a negative relationship between
PC3 and Air Temperature. It can be concluded that if the PC3 value increases,
the Air Temperature value will tend to decrease. PC4 has a coefficient
of – 0.482. This negative coefficient indicates that there is a negative
relationship between PC4 and Air Temperature. It can be concluded that if the
PC4 value increases, the Air Temperature value will tend to decrease. Apart from that, it is also
known that the intercept value found when the PC1, PC2,
PC3, and PC4 values are 0, Air Temperature is found to have
a positive value of 300.003.
In
this analysis, it is also known that
the MSE (Mean Squared Error) value
obtained is 0.909,
then the RMSE (Root Mean Squared Error) value is 0.953,
then the MAE (Mean Absolute Error) value is 0.806,
and the R-squared (R^{2}) 0.722.
2. Linear Regression Analysis without PCA
At
this stage the linear regression process will be carried out without
carrying out PCA analysis, where the 5 attributes will be retained
for linear regression analysis. In this stage, the Air
Temperature (Y) attribute will be used as the target or dependent variable,
while Rotational Speed (X1), Torque (X2), Tool Wear (X3), Process Temperature
(X4), and Machine Failure (X5) will be used as features or independent
variables. Details of attribute selection can be seen in Figure 14.
Figure 14. Selecting target
After
attribute selection is carried out, linear regression analysis can be carried out. The results of linear regression
can be seen in Figure 15.
Figure. 15. Result of linear
regression without PCA
From these results it is known that the linear
regression equation formed is:
𝑌 = −64.7421 − 0.00015771𝑋1 − 0.00518357𝑋2
− 3.46494𝑒05𝑋3 + 1.17797𝑋4
+ 0.828752𝑋_{5}
From
this equation it is known that Air
Temperature, which is the dependent variable, is influenced by 5 independent
variables, namely Rotational Speed, Torque, Tool Wear, Process Temperature, and
Machine Failure. Rotational Speed has a coefficient of -0.00015771. This negative
coefficient indicates that there is a negative relationship between
Rotational Speed and Air Temperature. It can be concluded that if
the Rotational Speed value increases, the Air Temperature value will tend to
decrease. Torque has a coefficient of -0.00518357. This negative coefficient
indicates that there is a negative relationship between Torque and Air Temperature. It can be concluded that if the Torque value increases, the Air Temperature value will tend to decrease. Tool Wear has a coefficient of -3.46494e05. This
negative coefficient indicates that there is a negative relationship between
Tool Wear and Air
Temperature. It can be concluded that if the Tool Wear
value increases, the Air Temperature value will tend to decrease. Process Temperature
has a coefficient of 1.17797.
The positive coefficient suggests a positive correlation between the air temperature and
the process temperature. It can be concluded that if the Process Temperature value increases, the Air Temperature value will also increase. The coefficient for machine failure
is 0.828752. The positive
coefficient suggests a positive correlation between Air Temperature and Machine
Failure. It can be concluded that if the Machine
Failure value increases, the Air Temperature
value will also increase. Apart from that, it is also known that the intercept value is found when the value of all independent variables is 0, so the Air Temperature is found to
have a negative value of -64.7421.
In
this analysis, it is also known that
the MSE (Mean Squared Error) value obtained is 0.909, then the RMSE (Root
Mean Squared Error) value is 0.953, then the MAE (Mean Absolute Error) value is 0.806, and the
R-squared (R^{2}) value is
0.772. Information regarding these results can be seen in Figure 16.
Figure. 16. Evaluation linear regression without
PCA
This
study utilizes linear regression analysis techniques to investigate predictive
maintenance. Two models were employed: one involved conducting principal
component analysis (PCA) before linear regression, while the other skipped PCA
and used the data directly. Surprisingly, both models yielded identical values
for key metrics such as mean squared error (MSE), root mean squared error
(RMSE), mean absolute error (MAE), and R-squared (R^{2}). This suggests
that the inclusion or exclusion of the PCA process did not significantly impact
the analysis, likely due to the characteristics of an artificial dataset
created by the author, which could be considered ideal. Moreover, PCA revealed
a high explained variance of 98%, retaining 4 out of 5 attributes that were
reduced, thus explaining the similarity in evaluation results between the two
models. Moving forward, future research should consider expanding the scope of
data incorporation, including relevant attributes, and ensuring the utilization
of up-to-date datasets. Exploring alternative analytical techniques such as
time series analysis and clustering may also offer deeper insights. Embracing
diverse analytical approaches is crucial to achieving a comprehensive
understanding of the investigated phenomenon. The author expresses gratitude to
the editor and reviewers for their invaluable feedback, which significantly
improved the paper's quality.
BIBLIOGRAPHY
Aalst,
W. V. (2016). Data Science in Action. Berlin: Springer. doi:https://doi.org/10.1007/978-3-662-49851-4_1.
Ebeling,
C. (2010). An Introduction to Reliability and Maintability Engineering. New York: MC Graw Hill.
Kherif, F., & Latypova, A. (2020). Chapter 12 -
Principal component analysis. Machine
Learning Methods and Applications to Brain Disorders,
209–225. doi:10.1016/B978-0-12-815739-8.00012-2.
Kiangala, K. S., & Wang,
Z. (2018). Initiating predictive maintenance for a conveyor motor in a bottling plant
using industry 4.0 concepts. International
Journal of Advanced Manufacturing Technology, 97(9- 12), 3251–3271.
Kim,
S.-G., Park, D., & Jung, J.-Y. (2021). Evaluation of One-Class Classifiers for Fault Detection: Mahalanobis Classifiers and the Mahalanobis–Taguchi System. Processes,
9.
Matzka, S. (2020). Explainable Artificial Intelligence for Predictive Maintenance Applications. 2020 Third
International Conference on Artificial
Intelligence for Industries (AI4I), (pp. 69-74).
Mobley,
R. K. (2002). An introduction to predictive maintenance (2 ed.). Amsterdam: Elsevier Science.
Montgomery, D. C., Peck,
E. A., & Vining, G. G. (2021).
Introduction to linear
regression analysis (6 ed.). New York: John Wiley & Sons, Inc.
Pranoto, J., Matondang, N., & Siregar, I. (2013). Implementasi Studi Preventive
Maintenance Fasilitas Produksi Dengan Metode Reliability Centered Maintenance Pada PT. XYZ. Jurnal Teknik Industri USU, 1(3), 18-24.
Richardson,
M. (2009). Principal component analysis. 6(16), 4. Retrieved from http://people. maths. ox. ac. uk/richardsonm/SignalProcPCA.pdf
Selcuk, S. (2017). Predictive maintenance, its implementation and latest trends. Proceedings of the
Institution of Mechanical Engineers, Part
B: Journal of Engineering Manufacture, 231(9), 1670–1679. doi:https://doi.org/10.1177/0954405415601640
Senanayaka, A., Mamun,
A. A., Bond, G., Tian, W., Wang,
H., Fuller, S., Bian, L.
(2022). Similarity-based Multi-source Transfer Learning Approach for Time Series Classification. International Journal of Prognostics and Health Management.
Wu, D., Jennings, C., Terpenny, J., & Kumara, S. (2016).
Cloud-based machine learning
for predictive analytics: Tool wear prediction in milling. In Proceedings - 2016 IEEE international conference on
big data, (pp. 2062-2069).
Zonta, T., Costa, C. A., Righi, R. d., Lima, M. J., Trindade,
E. S., & Li, G. P. (2020).
Predictive maintenance in the Industry
4.0: A systematic literature
review. Computers & Industrial Engineering.
Copyright
holder: Benedict Ariel Kurnianto, Rudy Hartono Prayogo,
Nidia Pialina Nababan, Suharjito (2024) |
First
publication right: Syntax Literate: Jurnal Ilmiah Indonesia |
This
article is licensed under: |