Predictive Maintenance using Machine Learning

A. Kane, A. Kore, A. Khandale, S. Nigade, P. P. Joshi

Computer Engineering Department, Pune Institute of Computer Technology

Abstract—Predictive maintenance (PdM) is a concept, which tubing machine.

is implemented to effectively manage maintenance plans of the
assets by predicting their failures with data driven techniques.
2) After analyzing the data set and deriving the correlation,
In these scenarios, data is collected over a certain period of time identify and train the most suitable Machine Learning
to monitor the state of equipment. The objective is to find some model which can predict the system failure well
correlations and patterns that can help predict and ultimately before in advance using the live data provided to the
prevent failures. Equipment in manufacturing industry are often corresponding model.
utilized without a planned maintenance approach. Such practise
frequently results in unexpected downtime, owing to certain 3) The model will be very beneficial for the operator
unexpected failures. In scheduled maintenance, the condition working on the machine and their respective managers
of the manufacturing equipment is checked after fixed time and other stakeholders in terms of operations. The
interval and if any fault occurs, the component is replaced to project will be very helpful in terms of saving time
avoid unexpected equipment stoppages. On the flip side, this
leads to increase in time for which machine is non-functioning and provide highly efficient results. Hence the overall
and cost of carrying out the maintenance. The emergence of productivity and profit increases.
Industry 4.0 and smart systems have led to increasing emphasis
on predictive maintenance (PdM) strategies that can reduce the
cost of downtime and increase the availability (utilization rate) II. L ITERATURE SURVEY
of manufacturing equipment. PdM also has the potential to
bring about new sustainable practices in manufacturing by fully 1) IJMOR 2020 Predictive Maintenance and Intelligent
utilizing the useful lives of components. Sensors in Smart Factory: Review. Said Boutahari,
Index Terms—Predictive Maintenance (PdM), Downtime Pre-
Mohamed Ramadany, Soukaina Sadiki, Maurizio Facci,
diction,Random Forest Regression, LSTM, Industry 4.O
Driss Amegouz The paper reviews papers concerning
predictive maintenance and intelligent sensors in
I. I NTRODUCTION smart factories. The paper’s main contribution is the
summary and overview of current trends in predictive
Downtime of the equipment is the most serious issue in
maintenance in smart factories.
the manufacturing industries. Failure of the equipment results
in downtime i.e. it remains useless until the maintenance is 2) IEEE 2019 Predictive Maintenance 4.0 P. Po´or, J.Basl,.
carried out and it starts functioning again. This scenario results Zenisek Talks about the various levels of predictive
in the decline of the productivity of the machine. maintenance. It also talks about the current stages
A common estimate is that almost every industry loses certain industries are at in the implementation of
5% to 20% of its productivity due to downtime and the predictive systems.
maintenance costs can be outrageous. Machine failures can 3) IEEE 2018 Forecasting faults of industrial equip-
bring production to a screeching halt, and it costs an industry ment using machine learning classifiers N. Kolokas, T.
millions of dollars.Reducing downtime can help in improving Vafeiadis, D. Ioannidis, D. Tzovaras Predictive main-
overall efficiency of the industry. tenance methodology to determine possible equipment
ML techniques help us by assessing the historical data of the stoppages along with the fault type in real time, uti-
equipment over a fixed period of time and develop some pat- lizing sensor data from operation times are discussed.
terns about degradation of the component. Thus we can predict The warning time frame is set by the plant operations
the potential failure and schedule the maintenance before the engineer to be as early as possible. Some widely used
system collapses entirely. This leads to minimum maintenance machine learning architectures are discussed.
cost, productivity is maintained and proper utilization of the
resources is ensured. III. OVERVIEW
We propose to develop a Web application which takes System failure is common occurrence across all the manu-
machine sensor data as input to predict possibility of downtime facturing industries. In most contexts, it is of utmost impor-
of the machine as a result of the entire chain of neural tance to predict the failure within time and take necessary
constructions. This will be done by steps.
1) Perform the data cleansing of the data set containing Many manufacturing industries face the problem of the dete-
the records of different parameters collected by the rioration of their equipment, which becomes critical especially
sensors of the tubing machine and find the correlation if this deterioration leads to a shutdown or unavailability of
to identify the pattern with respect to downtime of the that component. Therefore a concept called Preventative Main-
tenance (PM) emerged, which emphasizes on the prevention
of asset failures and their surrounding environments.
Predictive maintenance using Machine Learning techniques
tries to learn from data collected over a certain period of time
and use live data to identify certain patterns of system failure,
as opposed to conventional maintenance procedures relying on
the life cycle of machine parts.
The ML-based predictive approach analyses the live data
and tries to find out the correlation between certain parameters
to predict the system failure or schedule maintenance of the
equipment.
ML technology helps identify the fault lines by predicting
the failures at the right time and thus utilizing resources
Fig. 2. Dashboard
effectively. This ensures the establishment of balance between
maintenance needs and resource utilization.

IV. M ODULES
A. Server
The data from various sensors of the tubing machine
is fetched continuously and stored on the company server.
Company uses MongoDB database since the volume of data
generated is large. This data is then later used for the purpose
of maintenance of the manufacturing unit. This project deals
with small amount of data, hence uses MySQL database and
uploads the data on the server.
B. Client
The operator at a particular manufacturing plant can get
the updates of the manufacturing unit. He would get to know
the sensor readings at a particular timestamp from the server,
which would help him to know if the unit is functioning
properly or not, Hence, he can conduct the maintenance of the
machine accordingly, thus preventing the complete machine
failure.
C. Backend
The data fetched from the sensors is cleaned and pre-
processed to extract important features for data analysis and
finding patterns and correlations among the parameters. The
cleaned data is then used to train a machine learning model
which would predict the parameter values over a period of
time. Different models were trained on the parameters and
their accuracy was calculated. The models were trained using
classification and regression algorithms. LSTM model of deep
learning was also implemented to make the predictions.

Fig. 1. Methodological Steps Fig. 3. Parameters

 V. IMPLEMENTATION DETAILS
A. Dataset
The dataset used includes the sensor data of the machine
MNL15 of which the main parameters considered are men-
tioned below:
1) EJECTION-PCT: Excess material ejected from machine
during production.
2) EXTRUDER PRESSURE: Pressure sensor value from
the extruder of the machine.
3) MACHINE SPEED: Speed of the machine while man-
ufacturing.
4) ACTUAL-VALUES-INPUT: Input of raw material given
to machine.
5) HEATING ZONE: Temperature reading of each individ-
ual section. Fig. 5. Downtime Prediction by Model

B. LSTM
Long short-term memory is an artificial recurrent neural revolutions per min, ams/ccm input etc. These parameters
network (RNN) architecture used in the field of Natural are recorded by the sensors of the tubing machine over a
Language Processing and Deep Learning. Standard feed for- period of time, and hence maintenance can be scheduled as
ward neural networks such as CNN and RNN, don’t have per the requirements. This model will prove to be extremely
feedback connections. LSTMs have advantage in this aspect useful maintaining the productivity and minimizing the cost
over simple neural networks. LSTMs can process single data of maintenance.This will reduce the time for which the unit
points such as images as well as entire sequences of data such remains idle and maintenance. The model was very effective
as speeches. The standard LSTM unit is made up of a cell, in predicting the time when the machine will be down based
input gate, output gate and forget gate. The cell remembers on the previous historical sensor data that it received.
the numbers periodically and three gates control the entry
and exit of information from the cell. LSTMs have been REFERENCES
established to address the vanishing gradient problem that [1] T. Salunkhe, N. I. Jamadar, and S. B. Kivade, “Prediction of Remaining
Useful Life of Mechanical Components-A Review,” vol. 3, no. 6, pp.
can be encountered when training traditional RNNs. They can 125–135, 2018.
remember the information for long periods of time. [2] D. An, N. H. Kim, and J. H. Choi, “Practical options for selecting
datadriven or physics-based prognostics algorithms with reviews,” Reliab.
Eng. Syst. Saf., vol. 133, pp. 223–236, 2019.
[3] N. Sapankevych, R. Sankar, “Time series prediction using support vector
machines: A Survey”, 2009.
[4] Angius, A.; Colledani, M.; Yemane, A. Impact of condition based
maintenance policies on the service level of multi-stage manufacturing
systems. Control. Eng. Pract. 2018, 76, 65–78.
[5] Ribeiro, I.M.; Godina, R.; Pimentel, C.; Silva, F.J.G.; Matias, J.C.O.
Implementing TPM supported by 5S to improve the availability of an
automotive production line. Procedia Manuf. 2019, 38, 1574–1581.
[6] Lasisi, A.; Attoh-Okine, N. Principal components analysis and track
quality index: A machine learning approach. Transp. Res. Part. C Emerg.
Technol. 2018, 91, 230–248.

Fig. 4. Train vs Validation graph

VI. C ONCLUSION
This model proposes a system which will predict the failure
of the manufacturing unit based on parameters like tempera-
ture, pressure and also the machine parameters like speed,