International Journal of Recent Technology and Engineering (IJRTE)

ISSN: 2277-3878, Volume-8 Issue-4, November 2019

Role of Machine Learning in Manufacturing

Sector
Vimanyu Chopra, Devinder Priyadarshi

 between machines and human beings needed for greater

Abstract: This paper discusses the basic concept Machine efficiency and better performance. The various challenges for
learning and its techniques, algorithms as well the impact of employing machine learning in the field and the future scope
Machine Learning in Manufacturing processes and Industrial of the technology is also discussed.
Production. There has been an unprecedented increase in the data
available in the last couple of decades. This has enabled machine
II. TECHNIQUES
learning to be applied in various fields. Machine learning is field
of study which enables the computer system to learn automatically Machine learning uses many algorithms and tools for its
as well as improve from experience and perform various tasks implementation. Some of the types of machine learning along
without explicit instructions. The primary aim of machine with their algorithms are:
learning is to allow the computers learn automatically without
human assistance or intervention and adjust actions accordingly A. Supervised Learning
It is being employed in many fields. Manufacturing one area
Supervised machine learning algorithms use previous
where machine learning is very useful. This paper discusses the
various areas where machine learning can improve the process of
instances which are used to make prediction for future
manufacturing like predictive maintenance, process optimisation, instances. The data is divided into training and test dataset
quality control, scheduling of resources among others. This can [1]. A function is inferred from the training data which can be
be done by employing various machine learning techniques and used to make predictions about output values. This approach
algorithms using concepts such as deep learning, neural after sufficient training can be used to provide targets for new
networks, supervised, unsupervised and reinforcement learning. inputs ad can also compare the predicted output to the
The relationship of how machines and humans can co-exist and intended value of output which is useful in finding errors so
work together to improve the efficiency of production is also that the model can be modified accordingly. This approach
discussed. Industry 4.0 or fourth revolution that has occurred in
when training intends to find some patterns in training dataset
manufacturing which deals with advent of automation in
manufacturing industry and its significance is discussed. We take which can be used in prediction or classification [1]. Some of
a look at the various benefits and applications of machine the most popular algorithms in supervised learning are:
learning in the field of manufacturing engineering. This paper  Decision Trees
also discusses the various challenges and future scope of  Naive Bayes
employing machine learning in the manufacturing.
Decision Trees
Keywords: Machine learning, Industry 4.0, Manufacturing, Decision trees (Fig 1) are the type of trees which based on the
Automation. values group the attributes together. Decision trees rely
heavily on pre-determined classification [1]. Decision trees
I. INTRODUCTION are used in classification to identify the attributes which can
Machine learning as the name suggest is the ability of the be used to extract the most information required for solving
machine to learn by themselves. It deals with the programs the classification puzzle.
which learn over time to improve and adapt the performance
based on experience and past results without explicit
instructions from humans. Machine learning is being
employed in various fields ranging from image processing,
predictive analysis, military to medicinal field. It has
revolutionised the manufacturing industry over the last few
decades. There are many techniques and approaches which
enable the use of machine learning across various fields. This
paper first discusses these techniques employed in
manufacturing industry and then we take a look at the various
aspects of industry that machine learning has helped
including- predictive maintenance, quality control, Fig 1 - Decision Tree [19]
optimization and scheduling, supply chain forecasting as well Naive Bayes
as the concept of industry 4.0 and the symbiotic relationship Naive bayes is a classifier which is based on bayes theorem of
conditional probability.
Revised Manuscript Received on November 15, 2019
* Correspondence Author ∣
Vimanyu Chopra*, UG Scholar, DAV Institute of Engineering & P(c∣x) = P(c).
Technology, Jalandhar, India. Email: [email protected]
Devinder Priyadarshi, Faculty, DAV Institute of Engineering &
Technology, Jalandhar, India. Email: [email protected]

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2320 & Sciences Publication
 Role of Machine Learning in Manufacturing Sector

Fig 2. Example of Bayesian Network [18]

 When predictor (x, attributes) of class (c,target) is

given then posterior probability is donated by P(c|x)
 The class prior probability is denoted by P(c)
 P(x|c) is the likelihood which is the probability
of predictor given class. The probability of
predictor which is the likelihood is denoted by
P(x|c)
 The predictor prior probability is denoted by P(x)
It forms bayes network which is a group of trees based on
probability of things happening [2]. It is simple to apply and
is widely used for classification and clustering large datasets.
Disadvantage of using these classifiers is the requirement of
these predictors to be independent. It is used for real time Fig 3. Clustered Data [20]
prediction, text classification and sentiment analysis, Principal Component Analysis
recommendation systems.
Multivariate Analysis generally involves a large number of
B. Unsupervised Learning correlated values in the data. Principal Component Analysis
These techniques are applied to the dataset which has neither can be used to reduce the large amount of variables to a
labelled nor classified Unsupervised learning is followed smaller set while still containing the important information.
when input variables are known and we have no idea about This mathematical procedure converts correlated variables to
output variables. This approach is used to uncover unknown smaller sets known as principal components. For example a
patterns and find structures which might be hidden in the 2d graph can be represented as a 1d graph while still
datasets. The most common technique used is clustering in containing the important information with the help of PCA
which unorganized data is split into similarity groups. [2].
Clustering aims to form groups or “clusters” of similar data
based on metrics such as probabilistic distance [2]. Some
popular unsupervised learning models are:

 K means clustering
 Principal Component Analysis

K means clustering

K-means clustering aims to form clusters which are groups of

data points grouped together because of similarities. The first
Fig 4. Data visualization before and after applying PCA
step is to define the number of clusters k and the centre of
[1]
these clusters which can be chosen randomly. These
C. Semi Supervised Learning
centroids are placed as far away from each other as possible.
This mode of learning involves the power of both
The data points belonging to a data set are then associated to above-mentioned techniques - supervised and unsupervised
the nearest centre and keeping the centroid as small as learning [1]. This technique is utilised when a large amount
possible. At the end of this stage k is re-calculated and the
of unlabelled data is present along with a small amount of
same data points are allocated again to the new k centres [2].
labelled data in the dataset.
This process is repeated iteratively till there is no change in
the position of the centroid.

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2321 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-4, November 2019

This labelled data is used for training while also improving III. APPLICATIONS
accuracy. Some categories of semi-supervised learning are:
Machine learning is being widely utilized in various fields
 Generative Models
like healthcare, agriculture, education, automotive industry
 Self-Training etc. Some of the most notable uses of machine learning
 Transductive SVM include:
 Computer Vision
D. Neural Networks
Autonomous vehicles and self-driving cars utilise
Neural networks are derived from biological neural machine learning algorithms for extracting high level
networks. Neural network is not a machine learning
understanding from images and videos for making just
algorithm rather it’s a framework of many machine learning
like humans do by analysing these for making decisions
algorithms which work together to process the data. It works
identically to a biological neuron which consists of dendrites while driving.
which receive electrical signals, soma which process these  Virtual Assistants
signals and axon which take output to terminals [1]. Neural
Virtual assistants like Siri, Alexa, Google Assistant
networks contain three layers and the data flows
make use of machine learning for natural language
unidirectionally through these layers. The layers are:
processing while also collecting data and refining the
 Input layer
search results for a particular user according to their
 Hidden layer
searching habits.
 Output layer
 Medical Diagnosis and monitoring
Various tools based on machine learning are being
worked upon which are intended to provide basic
healthcare diagnoses based on historical data and
patterns. These tools form the diagnosis based on the
symptoms of the patient as well as information like
gender, age and previous history of illness in the patient.
This is especially useful in areas where medical
facilities are not available. Machine learning is also
being utilized in MRI scanning for identifying
abnormalities and anomalies in the scan as well as in
ICU for monitoring the patients.

 Product Recommendations
Fig 5 Artificial Neural Network [1] Many online stores and e-commerce websites nowadays
The input layer takes the input which is processed by the used machine learning models to determine the most
hidden layer and finally output layer is responsible for sends suitable product recommendations according to each
the calculated output [1]. They can follow all three modes of individual user based on their search and purchase
learning and can thus be further classified as:
history, increasing the likelihood of the customer to buy
the recommended product.
 Supervised Neural Network
 Unsupervised Neural Network  Facial, Speech and Handwriting recognition
 Reinforcement Neural Network These applications are possible because of machine
learning models which are trained extensively for these
Neural networks are popular because of their varied
purposed and are employed widely in various
applicability as they are used for various tasks like computer
vision, speech and handwriting recognition, robotics etc. [3]. applications and products.
Their drawback is the difficulty in understanding the outputs Machine learning is being incorporated in manufacturing and
along with time taken to train the model are being addressed production industry as well. The four major applications of
with the advancement of technology. machine learning in the field of manufacturing are:
E. Reinforcement learning
Reinforcement learning employs suitable action which is  Predictive Maintenance
required to achieve the maximum rewards[2]. It is used to  Supply Chain Forecasting
find the best path or behaviour which is required in a  Quality Control
particular environment or situation. Unlike supervised  Dynamic Scheduling
leaning the user has no idea about the outcome and thus this
model learns from experience to find the best course of action These techniques are discussed further in the following
to be taken for a positive result and the next decision to be section.
taken is based on previous one[4]. It employs delayed error
and trial and error search to get the ideal behaviour in a
situation for maximum efficiency.

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2322 & Sciences Publication
 Role of Machine Learning in Manufacturing Sector

Applications in Manufacturing example denotes the errors while the negative ones
The various applications of machine learning in the signify regular operation. The objective is to determine
manufacturing industry are: the probability of each upcoming example i.e. whether
it would fail or work normally within the next time
A. Predictive Maintenance
period.
In any major industry the reliability of machines and  Regression
equipment is very important as the unexpected breakdown of
these machines result in economic loss as well as halts the Regression is generally used when data exists in a range
production. The reactive maintenance cost is significantly and used to predict response for continuous values. In
higher than predictive maintenance cost [5].These predictive maintenance this method is used to determine
breakdowns can be avoided with help of predictive the Remaining Useful Life (RUL) of the equipment in
maintenance which continuously monitors the machines and question which is the time it would function properly
equipment as well as improves the system reliability. A before the next failure [7]. The remaining life of each of
significant amount of time is spent on locating source of the example as a continuous number is found. Linear
problem and with help of automated supervision system this regression is the most common method used due to its
time can be reduced and the production process can start speed and simplicity.
again. Automation of industry has brought about new A proactive maintenance approach is being considered which
technologies. The machines are equipped with various will enable us to monitor the machine adaptively and issue a
sensors which monitor the status of machine and the process warning before a breakdown can occur. This approach aims
being performed by it in real time. With data from these to eliminate failure equipment and increase
sensors we apply the machine learning algorithms to perform mean-time-before-failure to essentially infinity. This will
predictive maintenance to determine the state of machine and prolong the life of the machine to unlimited length [5].
predict its failure in the future [6].
B. Supply Chain Demand Forecasting
Generally, the data is collected form each individual devices Supply chain management is the management of the goods
and the data needs to be synchronised to get the information present in the system. It is very important in the industry and
about the whole system. Machine learning methods require includes storage of raw materials, inventory and includes all
data such as the fault history, maintenance history as well as the processes which are needed to get the final product.
the data about condition of the machine. These techniques try Collaboration among the partners in a firm and sharing
to predict the failures due to degradation of equipment, some information is an important part in reducing demand errors
unexpected breakdowns that might occur due to anomalies but this approach is not always feasible as the full
[7]. collaboration between stakeholders at all times is unfeasible
Methods for detecting faults are: [9]. Thus, supply chain forecasting is introduced to improve
 Modelling and parameter estimation approach the savings. Some of the critical issues required that permit
 Discrete observation approach supply chain collaboration to be successful are listed below:
 Finite automata state machine  Business interests should be similar
 PLC-based logic diagram  Management of relationship in the long-term
 Simulation  Unwillingness to share information
 Event-based time-window approach [5]  The convoluted design and management of large-scale
chains
The fault diagnosis is done based on the training the machine  The workers of supply chain should be competent [9]
learning models. These models also predict the lowest
downtime for maintenance by studying the peak hours and The machine learning algorithms can perform this
the work hours of the manufacturing industry. The data forecasting better than humans as the system is very complex.
collected is firstly pre-processed where it is made ready to be Some of the traditional forecasting methods are listed below.
input into the model. The information collected form the Naive forecasting
equipment is sorted into records and classified into categories Naïve method is the simples and is generally used as a
such as maintenance records, fault history, machine baseline to which other models are compared to. This method
conditions etc which is followed by the designing of takes the smallest value for future estimates [8].
functions and sequence aggregation. The predictive
maintenance tools mainly follow the following approaches Moving average
This method uses the average of some previous entries or
 Classification historical data entries for prediction[8].
Multiple Linear Regression
Classification is used to calculate the probability of This method uses previous demand changes to predict the
whether the machines or the equipment will fail over the upcoming changes in demand by treating the previous
course of a time period. This is determined with the help changes as independent variables. [8]The more advanced
of data acquired from the system[3]. Thus, the time for
methods act as universal approximators can approximate any
maintenance is also determined which is required to
avoid the downtime and breakdown of the machine. For function to an arbitrary accuracy. These approximators can
this approach two examples need to be identified which learn any function and thus all
are positive and negative. Each of these examples machine learning estimators
describes the operating conditions of the asset in
question with the help of historical records. The positive

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2323 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-4, November 2019

fall under this category of universal approximators thus machine learning algorithms provide a way in which
Neural Networks dispatch rules can be changed dynamically according to the
state of the system [13].The approaches to solve problems in
There are a number of neural networks which can be FMS scheduling can be categorized as
employed but the most popular is feed forward error  The analytical approach.
back-propagation type of networks. In these networks layers
 The heuristic approach.
of elements (neurons) are used. The output signal from one
 The simulation-based approach.
layer is transferred to the neurons of next layer. Thus, the data
 The artificial intelligence-based approach
moves layer to layer and additional layers called hidden
layers can be added between the input and output layers to The first approach views the flexible manufacturing system
increase the performance of these neural networks [8]. scheduling problem as an optimization model having certain
Support Vector Machines constraints. To resolve this model an appropriate model is
then chosen.
These machines are a kind of universal approximators which Heuristic approach is generally used to solve NP-complete
follow the structural risk minimisation principle as opposed type problems [13].
to empirical risk minimization principle followed by other Simulation based approach uses simulation for testing
methods[8]. SVM algorithms are used to reduce true errors dispatching rules and choosing the most efficient one.
on a random dataset by projecting the data into a higher Artificial intelligence-based approach uses techniques to
dataspace or hyperplane and increasing margins between dynamically change dispatching rules based on previous
classes and reducing error margins for regression. data.
C. Quality Control The machine learning model can be trained with the data
collected from previous runs of the production and
In today’s industry the competition is fierce and the company
techniques like inductive learning, case based learning or
wants to deliver the best possible product to the customer, but
neural networks can be applied to make the best possible
manufacturing process is prone to errors and a few out of
decision regarding the dispatch rules and the scheduling of
millions of products can be prone to errors. This can affect
resources in the production process to increase efficiency
the product and brand image in a negative way [10]. To
[13].
reduce this and to impose quality control machine learning
algorithms and techniques are being applied as the
IV. INDUSTRY 4.0
manufacturing processes are becoming more and more
complex and the conventional quality control techniques are Machine learning and AI has brought about a revolution in
not feasible[11]. The processes in a production line are linked the manufacturing industry where the traditional methods of
to one another and these processes being independent can production are being replaced by the new methods where
lead to variations and errors in a product [11]. Quality control smarter production is taking place as the machines are able to
is achieved by error detection as well as predictive quality take decisions and improve automatically by continuous
control. The data collected is used to predict the pattern of learning process, this revolution is generally known as
failures and find the exact factors which influence these Industry 4.0.As the industry now uses IOT and sensors which
supply chain problems by analysing large sets of data collect a huge amount of data every day ,this data is used to
collected by the company every day and also forecasting the find patterns and make predictions to improve the process of
supply demand with accuracy .The patterns in previous data production. This phenomenon uses various machine learning
can be used to predict the anomalies and detect the errors in techniques like – reinforcement learning, neural networks,
the given dataset like when system detects too much variation supervised learning among others [14].
in same product it triggers a response[11]. This also reduces One Important aspect of industry 4.0 is the symbiotic
the probability of occurrence of bullwhip effect which is the relationship between humans and machines. The machines
phenomenon in which creates false supply and demand are computationally faster they provide an advantage over
fluctuations. humans in terms of some processes like decision making
Quality assessment of products is one of the most important
and time-consuming tasks in the manufacturing industry,
now this intensive task of testing the products for errors is
performed by these algorithms. The quality of product can be
improved by optimizing the assembly line as well as
predictive maintenance of the machines. One such example
of quality control using machine learning is the application of
supervised learning and clustering to the process states as
well as deep learning.
D. Dynamic Scheduling
Scheduling is a very important part of production process as
when there are resources being shared by multiple entities it
is very important to schedule these resources efficiently so as
to improve the performance of the system [12]. Dispatching
rules are used to schedule tasks in a flexible manufacturing
system (FMS) and it is very difficult to change these dispatch
rules according to changing situations by traditional methods,

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2324 & Sciences Publication
 Role of Machine Learning in Manufacturing Sector

Table- I: Comparison of humans and machines [15]

Factor Humans Machines
The differences between individuals are
Very low
high
Inadequate maintenance can
Training and employee satisfaction can
result in degradation during its
decrease Quality variation and improve
Mechanical Jobs lifetime
quality
Disparity of Moderate based upon factors
Problem solving abilities yield vast
Quality in such as data quality, precision
differences between individuals
of algorithm etc.
It can be ameliorated by using
Interests (like personal, institutional) may
large datasets to train the
impact this process.
Decision Making machines.
Decision Complexity and other factors These factors do not affect
may affect the decision-making process machines
While a task is High Low
Variation in
being High likelihood of dissatisfaction and
Performance Negligible
performed/executed weariness from work
during ramp up process. Humans on the other hand while
being at a disadvantage in some scenarios like iterative
processes are sometimes more efficient in making decisions V. CHALLENGES
based on experience and in situations with incomplete data.
Machine learning apart from manufacturing industry is now
So, there is a need for a human-machine relationship in which
being employed in almost every major filed like healthcare,
both the entities work together to provide maximum
education, military among others. With the advent of
efficiency. It has been observed that a human guided
exploration in ramp up processes was more efficient than the Industry 4.0 the role of machine learning algorithms is
traditional machine learning approaches receiving close to increasing day by day as they provide accurate predictions
and improv the efficiency of production processes in addition
optimal behaviour by combining best of each of their abilities
to improving the cost savings of companies.
[16].
For a machine learning technique to be viable to be used in
One such example is where a human designs work plans the industry, the following properties are required -
which are product specific and a sequencer in a factory which  Ability to handle various data types (text, numbers,
is responsible for scheduling of tasks and it obtains its etc)
knowledge of task plans and the order of jobs from  Ability to handle noise, and fuzzy data
information of these work plans. Humans are not able to keep  Processing should be done in real-time
up with the changes required due to situational changes. In a  Ability to handle high dimensional and large data
human-machine relationship both can benefit as the humans sets
can gain an understanding from the sequencer
 Easy to comprehend results should be generated
recommendations/suggestions and decisions and on the other
 Simple implementation [3]
hand when humans overrule the decision the machine can
Implementation of these machine learning models in the
learn decision making behaviour [15].
industry faces many challenges. Some of the challenges
Another example might be the sequencing of the tasks to be which are widely accepted are:
performed during production. The sequence of tasks is  Using modern technologies in manufacturing.
determined by a sequencer which assigns times to different  Increasing emphasis on producing high-value added
tasks. Generally, the rule of first-in first-out is followed but commodities
the sequencer also changes the sequence dynamically if some  Using artificial intelligence, advance knowledge
changes occur in the production process. The sequencer  Products to be manufactured and the process to be
utilises the data based on the work plan prepared by a human employed should be feasible.
worker. This provides us with best of both worlds as humans
 Improvement in services, products.
cannot change sequence of tasks dynamically as quickly as
 Increased cooperation between researchers and
the machines. Thus, with the guiding input from humans the
manufacturing industry regarding latest
sequencer performs at peak efficiency increasing resulting in
technologies.
a smooth production process.
 Modern manufacturing management standards [17].
Thus, the respective capabilities of humans and machines are A major challenge in implementing these techniques is the
mutually beneficial to one another as humans are better at gathering of data or specifically usable data as the data
analysing and countering unexpected errors and new generally acquired contains large amount of redundant data
situations whereas machines are better at iterative tasks and which can affect the output.
pattern recognition. Another challenge is the

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2325 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-4, November 2019

Table- II: Suitability of ML techniques in process of manufacturing [17]

Manufacturing requirement Ability (Theoretical) of Machine Learning to meet the

requirements
The system should have the potential to reduce complicated Machine learning enables automatic recognition of the
results and present accurate and clear advice to users existing patterns an thereby deduce algorithms for
upcoming events. This enhances the decision-making
ability of the process owners.
The system should have the potential to dynamically change As ML is based on AI, it thus has the ability to
according to environmental conditions keeping in check the adjust/modify dynamically according to the requirement
effort and cost involved without involving the system operator.
By studying the results, the system should have the potential By seeking out patterns in data, machine learning are used
to improve the pre-existing knowledge to derive new information or even knowledge.
Even without special requirements at the beginning the Machine learning techniques employ already stored data
system should have the potential to use the manufacturing to convert it into information which can be used in various
data available to perform the tasks. applications like making predictions in a system
The inter and intra relations should be found and correlated Detection of irregularities and patterns which
by the system define data is the main aim of some machine learning
approaches
Ability of handling high-dimensional problems and datasets Some problems like over-fitting need to be taken care of
even in machine learning approaches which have the
ability to handle data of high dimensionality

pre-processing which can have problems like unbalanced or include it as it may become relevant under certain
incomplete/incorrect data set values and adding or replacing circumstances down the road [17].
values can influence the results. Another hurdle is to select
the best machine learning algorithm suited for the situation VII. ADVANTAGES
among a large number of such algorithms as there are a
number of factors which need to be addressed while choosing Machine learning approaches are used to solve optimization
the algorithm which would be best suited to the problem. problems in intelligent manufacturing system as they are
competent enough to tackle NP complete problems. Most of
General approach for selection of algorithm consists of three
the problems encountered in manufacturing are data rich,
steps-
machine learning helps to improve understanding of data by
1. The data is analysed to see whether it is labelled or
extracting implicit relationships from large data-sets due to
unlabeled to choose between the best approach –
its capability of handling data of high dimensions and
supervised, unsupervised or reinforcement learning multi-variate datasets. Machine learning is universally
2. Checking the applicability of current algorithms believed to improve utilization of resources and reduces
based on various factors like data types, amount of cycle time and scrap in some of the NP-hard problems. In
data available etc. complex processes like manufacturing of semiconductors it
3. Investigation of algorithms in similar cases to provides tools for continuous quality control and
identify the best algorithm for the given problem. management of production process.

Nowadays a combination or hybrid technique is also used As mentioned earlier, machine learning algorithms are
which use more than one algorithm to tackle the problem equipped to handle problems and data of high dimensions but
with better accuracy and results than a single algorithm. to varying degrees depending upon the algorithms or
Security of data is also a major concern as most data in this technique chosen and cannot be generalised. Some
age is being stored on cloud platforms making this data algorithms like Distributed Hierarchical Decision Tree are
vulnerable and it is possible to gain remote access to these known to handle the high-dimensional data and problems
machines if appropriate security measures are not taken [14]. better than other and are thus preferred in manufacturing
Lastly, the interpretation of results acquired from machine process as large amount of data encountered is of high
learning techniques is very difficult along with difficulties in dimensionality.
finding the talent for such jobs as this is a relatively new field
and finding qualified individuals is a difficult [17]. Various tools and programs are now available nowadays for
application of machine learning algorithms and has made
VI. SUITABILITY OF MACHINE LEARNING ALGORITHMS process easier offering features like adjustment of parameters
Many machine learning algorithms handle large amounts of easily in order to improve classification performance.
data and one of the problems they might face while handling
data with high-dimensionality is over-fitting. This problem We prefer of machine learning algorithms due to their ability
rarely occurs due to power of modern algorithms but there of identifying implicit relationships in large and chaotic
exist techniques to reduce dimensions should the problem data-sets by using Supervised, Unsupervised or
occur. In techniques like Support Vector Machines the need
for reducing dimensions is reduced as dimensionality is not a
practical problem. This in turn means that we can be more
liberal towards information which may seem irrelevant and

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2326 & Sciences Publication
 Role of Machine Learning in Manufacturing Sector

Reinforcement learning depending upon data-set which was diversified and accurate. This means that the individual
not possible until a few years ago. Thus, the huge amount of classifiers should be accurate meaning they should provide
data collected over the decades can now be utilised for results which are an improvement over guessing and thus
improving performance and production process. they should which should result in varying errors when
encountering new data points.
Another advantage of machine learning algorithms is their There are three approaches for generating multiple classifiers
ability to adapt dynamically to the situation. Manufacturing Bagging
and production processes are very uncertain and problems
can arise at any time for instance breaking down of a machine In bagging technique, a replication is generated of the
or bottleneck in some machines. Machine learning allows the training set by sampling with replacement from the training
production process to run smoothly by techniques like instances [3] with each training set having same size as that of
continuous monitoring of machines, dynamically changing the original data. Examples may appear multiple times or not
sequence of jobs scheduled adapting to change in appear at all. These types of training sets are defined as
circumstances in production or scheduling repair and bootstrap replicate of original training set with technique
maintenance of machines during off hours or when the being called bootstrap aggregating. A training classifier is
machine is not being utilised. This enables production line to derived from each replication of training set and all the
run smoothly without any interruption leading to an increased classifiers generated then use voting scheme to classify each
production rate which may not be achievable by traditional instance in test case.
methods and practices. [17] Boosting

In this technique boosting maintains for each instance in the

VIII. CURRENT TRENDS AND FUTURE OF MACHINE
training set a weight which signifies its importance instead of
LEARNING IN MANUFACTURING
drawing bootstrap samples from the original instance
Machine learning is experiencing rapid progress in various successively. The higher the weight of an instance, the more
fields including manufacturing and industrial production. it influences the learned classifier. The weights are adjusted
The two most important directions taken nowadays in the during each round according to the performance of
field of machine learning are: corresponding classifier. A weighted voting scheme where
Scaling up of machine learning algorithms every component classifier contributes with different strength
During the last couple of decades there has been a drastic based on their accuracy of training is used to construct the
increase in the amount of data collected in every field final classifier.
including the manufacturing industry. There is large amount Error-Correcting Output Coding (ECOC)
of data collected during the production process in every step
from production planning, design to maintenance. Scaling up This method was devised to solve multiple two-class problem
of machine learning algorithms is thus required to as a means of handling multiclass problems. In
accommodate and process these large data-sets in the best Error-Correcting Output Coding each binary classification
way possible. Data mining which is an field which uses these function corresponding to a unique partitioning of class is
algorithms has highlighted this demand as it requires encoded by a set of output bits of classes. Class predictions
processing millions of training examples and handling are then generated by combining all functions which is
terabytes of data [3]. derived from algorithms that use ECOC [3].
This scaling of algorithms can be done by two approaches
Machine learning has been employed by many
 By producing more efficient algorithms Multi-National Companies in their production plants around
the world. However, the machine learning field has a lot of
 By partitioning of data
scope for improvement as it is a relatively new field. There
In the first approach the aim is to generate machine learning are many areas which need further research to improve and
algorithms which are more efficient or improve the optimize the techniques we employ for better results. Many
performance of current algorithms to improve the processing of these are already being worked upon. Some of these areas
time and performance of machine learning model. This are:
includes techniques like optimizing search an representation,  Utilizing existing knowledge
finding approximate result instead of an exact one.  Learning in structural domains
In the second approach the data set is broken up into subsets  Learning with continuous variables
which avoid running the algorithms on very large sets of data.  Constructive / meta learning
Generally, data is divided into subsets with common features  Integration of machine learning with existing
to make processing simple. This approach also aims to avoid
engineering software
memory management issues [3].
Learning Multiple Models  Exploratory systems [21]

This is another area currently being heavily researched. This IX. CONCLUSION
approach involves increasing accuracy in supervised
learning. This is achieved through using ensemble of Machine learning has become a very important part of the
classifiers or multiple models. These refer to combining manufacturing industry in recent times and its importance
classifiers in some fashion rather than using them will increase further in the
individually. It is believed that ensemble of classifiers are an future.
improvement over individual classifiers in terms of accuracy,
provided that the individual classifiers are themselves

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2327 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-4, November 2019

It has brought about the industry 4.0 revolution and is 17. T. Wuest, D. Weimer, C. Irgens, K.D. Thoben, "Machine learning in
manufacturing: advantages, challenges, and applications." Production
changing the face of production process with techniques like
& Manufacturing Research, vol. 4, no. 1, 2016, pp. 23-45.
neural networks. The manufacturing process has become 18. R. Gangula, G. Nandam, C. Sudha, S. Rekha, “Usage of Machine
more efficient and the production line more streamlined with Learning Algorithms in Data Mining”, International Journal of Recent
machine learning various applications like predictive Technology and Engineering (IJRTE), vol.8, no. 1C2, 2019, pp.
411-414
maintenance, scheduling, quality control etc. The symbiotic
19. L. Rokach and O. Maimon, "Top-down induction of decision trees
relationship between humans and machines is very important classifiers - a survey," IEEE Transactions on Systems, Man, and
as it results in near optimal performance. The development of Cybernetics, Part C (Applications and Reviews), vol. 35, no. 4, 2005,
machine learning algorithms is advancing at a rapid rate and pp. 476-487.
20. http://pypr.sourceforge.net/kmeans.html (Accessed on oct. 12, 2019)
along with the increase in computational power, the
21. B.L. Whitehall, S.CY. Lu, "Machine learning in engineering
challenges like interpretation of output and data and automation—The present and the future.", Computers in Industry,
pre-processing can be solved in the future leading to vol. 17, no. 2-3, 1991, pp. 91-100.
industry-wide implementation of machine learning
techniques for improved efficiency and production. AUTHORS PROFILE
Vimanyu Chopra is an undergraduate scholar at
REFERENCES DAV Institute of Engineering and Technology,
Jalandhar, Punjab, India.
1. A. Dey, "Machine learning algorithms: a review.", International
Journal of Computer Science and Information Technologies, vol .7,
no. 3, 2016, pp. 1174-1179.
2. T. O. Ayodele, "Types of machine learning algorithms.", New
advances in machine learning. IntechOpen, 2010, pp.19-48
3. D. T. Pham, A. A. Afify, “Machine-Learning Techniques and Their
Applications in Manufacturing.” Proceedings of the Institution of
Devinder Priyadarshi is Assistant Professor in
Mechanical Engineers, Part B: Journal of Engineering Manufacture,
Department of Mechanical Engineering at DAV
vol. 219, no. 5, May 2005, pp. 395–412.
Institute of Engineering and Technology,
4. L.Monostori, A.Markus, H.Van Brussel, E.Westkämpfer, "Machine
Jalandhar, Punjab, India.
learning approaches to manufacturing." CIRP annals, vol. 45, no. 2,
1996, pp. 675-712.
5. J.Lee, "Machine performance monitoring and proactive maintenance
in computer-integrated manufacturing: review and
perspective.", International Journal of computer integrated
manufacturing, vol. 8, no. 5, 1995, pp. 370-380.
6. A. Kanawaday and A. Sane, "Machine learning for predictive
maintenance of industrial machines using IoT sensor data," 2017 8th
IEEE International Conference on Software Engineering and Service
Science (ICSESS), Beijing, 2017, pp. 87-90.
7. G.A. Susto, J. Wan, S. Pampuri, M. Zanon, A.B. Johnston, P.G.
O'Hara, S. McLoone,"An adaptive machine learning decision system
for flexible predictive maintenance," 2014 IEEE International
Conference on Automation Science and Engineering (CASE), Taipei,
2014, pp. 806-811.
8. R. Carbonneau, K. Laframboise, R. Vahidov, “Application of
machine learning techniques for supply chain demand
forecasting”. European Journal of Operational Research, vol.184, no.
3, 2008, pp.1140-1154.
9. R. Carbonneau, R. Vahidov, K. Laframboise. "Machine
Learning-Based Demand forecasting in supply chains." International
Journal of Intelligent Information Technologies (IJIIT), vol. 3, no. 4,
2007, pp. 40-57.
10. C. A. Escobar, R. Morales-Menendez, “Machine learning techniques
for quality control in high conformance manufacturing
environment”. Advances in Mechanical Engineering, vol. 10, no. 2,
2018, DOI: 10.1177/1687814018755519.
11. T. Wuest, C. Irgens, K.D. Thoben, "An approach to monitoring
quality in manufacturing using supervised machine learning on
product state data.", Journal of Intelligent Manufacturing, vol. 25, no.
5, 2014, pp. 1167-1180.
12. P. Priore, D. de la Fuente García, I.F. Quesada, " Manufacturing
Systems Scheduling through Machine Learning”, NC, 1998, pp.
914-917.
13. P. Priore, D. de la Fuente, A. Gomez, J. Puente, “A review of machine
learning in dynamic scheduling of flexible manufacturing systems”,
Artificial Intelligence for Engineering Design, Analysis and
Manufacturing, vol. 15, no. 3, 2001, pp.251-263.
14. B. Bajic, I. Cosic, M. Lazarevic, N. Sremcev, A. Rikalovic, “Machine
Learning Techniques for Smart Manufacturing: Applications and
Challenges in Industry 4.0”, Department of Industrial Engineering
and Management Novi Sad, Serbia, 2018, pp.29.
15. F. Ansari, S. Erol, W. Sihn, "Rethinking Human-Machine Learning in
Industry 4.0: How Does the Paradigm Shift Treat the Role of Human
Learning?" Procedia Manufacturing, vol. 23, 2018, pp.117-122.
16. S. Doltsinis, P. Ferreira, N. Lohse, “A symbiotic human–machine
learning approach for production ramp-up.” IEEE Transactions on
Human-Machine Systems, vol. 48, no. 3, 2017, pp.229-240.

Retrieval Number: D8191118419/2019©BEIESP Published By:

DOI:10.35940/ijrte.D8191.118419 Blue Eyes Intelligence Engineering
2328 & Sciences Publication