© 2021 JETIR April 2021, Volume 8, Issue 4 www.jetir.

org (ISSN-2349-5162)

Machine Failure Prediction Using Machine

Learning

Varshini Manda, B.Tech Student, Department of Information Technology,

Dr. K. Neeraja, Professor, Department of Information Technology,

MLR Institute of Technology, Telangana, India.

Abstract-Industrial equipment performance However, finding relationships between

control and failure prediction are important not quantifiable software properties and faults [11,
just for the quality of the produced material, but 12] will aid in fault detection. Traditional
also for the amount of time and money saved in approaches, such as testing or simulation, are
overall maintenance. This project aims to monitor unable to meet the challenges of fault prediction
the evolution of AI/ML techniques for equipment [13-15].
fault prediction in industries over time. The topics
High cost and time overhead are two critical
covered in this paper include machine learning
considerations in this regard. Furthermore,
algorithms, use cases, and principles related to the
reenactment is unsuitable for promoting transient
application of such technology in a variety of
properties because it does not take into account all
industries such as software and hardware. This
of the framework's potential states. A formal
survey looks at early research from the late 1980s
method can be used to solve this problem.
to the early 2000s, as well as recent research from
Mathematical logic underpins formal methods.
the early 2000s to 2017, and the most recent
Formal methods are divided into two categories:
research from the last two years. It can be inferred
formal specification and formal verification [16].
that this project offers a detailed review of various
Formal specifications [17, 18] describe the
machine learning and artificial intelligence
interactions between fault proneness and formal
(ML/AI) approaches used in the Industrial
verification. Most fault prediction papers use
Manufacturing domain. LSTM was discovered to
simulation and experiments to test their proposed
be one of the most commonly used processes.
system. Model testing [19-21] is another method
1. INTRODUCTION for verifying an information-centric IoT app.

1.1 Introduction In this paper, fault behaviors are divided into two
The quality of Internet of Things (IoT) categories: dimension reduction and fault
applications [1] has improved dramatically in prediction, both of which can be demonstrated
recent years, and related problems have become using behavioral models [22, 23]. Deep learning
increasingly important for software developers [2, has been suggested as a result of the ongoing
3]. The capability of assessing the fault prediction growth of machine learning. Deep learning uses
process, evaluating the degree of a product multi-layer information processing and feature
module, and then testing [4, 5 Testing assists the extraction to estimate complex nonlinear
developer in reducing costs, and fault detection functions with small errors. In the literatures [11-
provides input for maintenance procedures [6, 7]. 12], convolutional neural networks and recurrent
At the programming process, programming neural networks have been suggested.
weaknesses are difficult to assess [8-10].
JETIR2104196 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 738
© 2021 JETIR April 2021, Volume 8, Issue 4 www.jetir.org (ISSN-2349-5162)

In the field of deep learning, they are of practical 1.2Problem Statement

importance. In time series prediction, recurrent • Equipment failure due to software issues, lack of
neural networks are more accurate. Deep training, use of AI/ML techniques on existing data to
on the other hand, will cause the neural network's predict failures, and performance deterioration of
gradient to vanish. On the basis of the recurrent industrial equipment are some of the problems and
neural network, the long short-time memory challenges faced in this domain.
network proposed in [13] adds three "gate"
structures and has a new improvement in time
series prediction with reasonably high accuracy.
OBJECTIVE:
The performance estimation of power plant
equipment is often influenced by a large number • In recent years, manufacturers have been looking
of multi-dimensional variables. When dealing at implementing newer machine learning
with a large amount of multidimensional data, approaches for predictive maintenance and
Long short-term memory networks are inefficient, equipment tracking. It is feasible to try building
and precision is sometimes lacking. Although a these sophisticated systems due to the availability
convolutional neural network can typically handle and affordability of powerful computing tools,
multi-dimensional data, the data's timing is not vast volumes of industrial data sets from their own
well reflected. Traditional particle swarms, plants, and deep-learning AI/ML algorithms.
attention systems, and other algorithms can be
used to process multidimensional data, resulting in
data reduction. 2. Literature Survey
These approaches also neglect the relationship Finding faults and assigning them to the
between input data, resulting in redundancy and appropriate developer to fix is critical in IoT-based
data loss. This paper proposes a model based on applications that are growing every day. The
CNN-LSTM applied to the equipment's fault following are three key directions that previous
prediction stage in order to find an appropriate studies [24] on IOT-based applications have taken:
solution. The convolutional neural network
component is especially effective at data 1) defining and specifying metrics for calculating
processing. It has the ability to reduce the amount software complexity,
of data needed for experiments. without
jeopardising the data's link, the extracted feature 2) validating correctness and thoroughness, and
vector is fed into a long short-term memory 3) identifying and investigating models that
network that excels at forecasting time series data. attempt to predict faults based on the software
This model guarantees not only the characteristics metrics specified [25, 26].
of the input data and their relationships, but also
the timing of the output data. Software metrics, which provide quantitative
images of programme qualities, can be used to
a model Finally, the functional value, model describe fault prediction in software. Several
prediction value, and LSTM network prediction studies show that programme metrics are linked to
value are compared. An example is used to fault proneness [27]. To construct predictive
demonstrate the method's logic and effectiveness. models of the future, a few techniques have been
The approach is also shown to have a greater investigated. Computer fault prediction Statistical
improvement impact than LSTM network methods [28] have also been proposed. Much
prediction. research has gone into determining the best way to
choose programme metrics that will most likely
reveal fault-proneness [29]. Some metrics for
JETIR2104196 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 739
© 2021 JETIR April 2021, Volume 8, Issue 4 www.jetir.org (ISSN-2349-5162)

describing software quality in static and dynamic computer defects as a result of ageing A tool for
platforms have been introduced. The features of predicting software ageing was proposed by Liu
code structure are calculated as metrics in the static and Meng [45]. The approach is based on a Back
platform [30]. A number of supervisors [31, 32] Propagation (BP) error neural network. The
and a number of bunches [33] are static Artificial Bee Colony (ABC) algorithm is used to
measurements. Testing perfectionism is measured determine the weights and thresholds. To put it
using dynamic platforms. Basic element another way, ABC is used to improve the BP
measurements are influenced by the spectrum of model. They demonstrated that, when compared to
auxiliary and information streams [6]. The the conventional BP neural network, their
relationship between product measurements and proposed approach converges more quickly and
blame proclivity, as well as a number of accurately predicts.
quantifiable programming characteristics, has
been precisely defined. Many researchers have
demonstrated this [22, 34-37]. Assurance of fault 3. OVERVIEW OF THE SYSTEM
prediction modules is important since it defines the
modules require itemized testing and replication In systems engineering, information systems, and
[38]. Measurements are used in early fault software engineering, the Systems Development
prediction algorithms [39]. Machine learning Life Cycle (SDLC) is the process of developing or
techniques [40], such as Support Vector Machine altering systems, as well as the models and
(SVM) [41], have gained a lot of attention methodologies that people use to build these
recently. systems. The SDLC definition underpins a number
of software development methodologies in
Artificial Neural Network (ANN) [42] and Naive software engineering.
Bayes algorithm (NB) [22]. The Multi-Layer
Perceptron (MLP) is a common supervised 3.1 Existing System
learning machine learning algorithm. The Particle  The following are three key paths that
Swarm Optimization (PSO) algorithm is an previous studies on IoT-based applications
optimization algorithm that uses evolutionary have taken: 1) identifying and determining
algorithms. Rui et al. [43] suggested a fault metrics for measuring machine
detection system for power IoT devices. complexity, 2) validating correctness and
Via deep learning, they proposed a multi-spectral thoroughness Some metrics for describing
approach for fusing images. A deep learning software quality in static and dynamic
convolutional neural network is used to detect platforms have been presented. The
faults in images of power devices. Their proposed features of code structure are calculated as
method aids in the accurate and fast location of metrics in the static platform. A number of
fault points. [44] suggests a hybrid solution supervisors and a number of bunches are
incorporating the Extreme Learning Process used in static calculations. Testing
(ELM) and the Genetic Algorithm (GA). With perfectionism is measured using dynamic
extracting useful information from the fault platforms. Measurements of basic
reports, a vector space model is constructed based elements are based on scope of auxiliary
on the information and a minimal feature set is and information sources Many researchers
selected. The features are fed into an ensemble have demonstrated the connection between
classifier that is a GA-based ELM training product measurements and blame
algorithm. Their proposed algorithm outperformed tendency, as well as numerous quantifiable
KNN, Naïve Bayes and SVM. Another issue in programming characteristics.
IoT environments is occurring
JETIR2104196 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 740
 © 2021 JETIR April 2021, Volume 8, Issue 4 www.jetir.org (ISSN-2349-5162)

3.1.1 Disadvantages of Existing System project.

 For the dimensionality process and Pre-processing:
prediction approach, existing methods are In this stage data set features are taken as
used. input and difference in time series data is
 Existing approaches are focused on calculated and new dataset is created which has
research protocols that can't predict abrupt difference values with previous values. This
data changes. dataset is used for predicting next values based on
difference as feature and next value as label.

3.1.2 Proposed System

Prediction Model:
 Identify potential computer failure early LSTM model is initialized for training
and warn the device. The approach is to use time series data. Using fitting function features and
a Deep Learning LSTM model to forecast labels are given as input and algorithm model is
time series values in advance, and then use trained for predicting future values.
a classifier to assess whether or not to issue Regression Model:
an alert. In this module data set with memory
values are used as features and alert or non-alert is
used as label. And model is trained and this model
 Identify potential computer failure early
is used for prediction failure condition.
and warn the device. The approach is to use
a Deep Learning LSTM model to forecast Prediction:
time series values in advance, and then use For prediction new dataset is taken with
a classifier to assess whether or not to issue 100 memory values as features and given input for
an alert. predict function of LSTM model and linear
 In this proposed scheme, the LSTM regression model and next value and failure
algorithm is used to create a model from a prediction value is calculated.
data set of an equipment. Values are
trained using a machine learning model, ARCHITECTURE:
which is then used to track fault detection
and generate warnings.
3.1.3 Advantages of Proposed System
 Machine learning methods are used to
detect faults in automated machinery
processes.
 There is no need for manual intervention or
checking.
 Processing time is cut in half.

4. CONCLUSION
3.2 System Modules
A short-term prediction approach based on the
Dataset Collection: LSTM network is proposed in this project, based
on existing fault predicting technologies. This
Machine data set with one feature and one
approach increases the accuracy of fault detection
label is used in this dataset. Feature with memory
and label with o and 1 are used as dataset in this
JETIR2104196 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 741
© 2021 JETIR April 2021, Volume 8, Issue 4 www.jetir.org (ISSN-2349-5162)

as compared to the CNN method. The benefit of of web software faults in a fuzzy
this approach is that it can predict the next value. environment using the MODULO-M
System malfunction is predicted and a warning is multivariate overlapping fuzzy clustering
displayed by comparing previous dataset values algorithm and a newly proposed updated
and based on the next value using a Logistic prediction algorithm," Appl. Soft Comput.,
regression classifier. vol. 22, no. 3, pp. 372-396, 2014.
 [8] C. Jin and S.-W. Jin, "Software
5. REFERENCES faultproneness prediction using a hybrid
artificial neural network and quantum
 [1] Souri, A. Hussien, M. Hoseyninezhad, particle swarm optimization," Applied Soft
and M. Norouzi, "A comprehensive Computing, vol. 35, no. 10, pp. 717-725,
analysis of IoT connectivity techniques for 10/ 2015.
an effective smart world," Transactions on
 [9] P. J. Garca Nieto, E. Garca-Gonzalo, F.
Emerging Telecommunications
Sánchez Lasheras, and F. J. de Cos Juez,
Technologies, vol. 11, no. 3, p. e3736,
"Hybrid PSO–SVM-based approach for
2019.
predicting the remaining useful life for
 [2] R. Mahajan, S. K. Gupta, and R. K. aircraft engines and assessing its
Bedi, "Design of Software Fault Prediction reliability," Reliability Engineering &
Model Using BR Technique," Procedia System Safety, vol. 138, pp. 219-231, 6/
Computer Science, vol. 46, 2015/01/01, 2015.
pp. 849-858.
 [10] V. Balasubramanian, F. Zaman, M.
 [3] In Proceedings of the International Aloqaily, I. Al Ridhawi, Y. Jararweh, and
Conference on Data Engineering and H. B. Salameh, "A mobility management
Communication Technology, 2017, pp. architecture for smooth delivery of 5G-IoT
779-787, I. Umesh and G. N. Srinivasan, services," IEEE International Conference
"Dynamic Software Aging Detection- on Communications (ICC), 2019, pp. 1-
Based Fault Tolerant Software
Rejuvenation Model for Virtualized
World."
 [4] Y. Abdi, S. Parsa, and Y. Seyfari, "A
hybrid one-class rule learning method for
machine fault prediction based on swarm
intelligence," Innov. Syst. Softw. Eng.,
vol. 11, no. 2, pp. 289-301, 2015.
 [5] M. R. Mesbahi, A. M. Rahmani, and M.
Hosseinzadeh, "Reliability and high
availability in cloud computing
environments: a guide roadmap," vol.8, p.
20, 2018.
 [6] I. Alsmadi and H. Najadat, "Evaluating
the adjustment of software fault behaviour
with dataset attributes dependent on
categorical correlation," Advances in
Engineering Software, vol. 42, no. 8/ 2011,
pp. 535- 546.
 [7] S. Chatterjee and A. Roy, "Prediction

JETIR2104196 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 742