Vibration Analysis for condition Monitoring & Predictive Maintenance using Embedded TinyML
0 likes129 views
This document describes a project to develop a predictive maintenance system for industrial equipment using machine learning. The system uses an accelerometer and Arduino board to collect vibration data from equipment. The data is used to train a machine learning model to detect anomalies indicating potential failures. When deployed, the model makes predictions in real-time which are displayed on a mobile app along with sensor data, allowing users to remotely monitor equipment condition. The goal is to help improve equipment reliability and availability while reducing maintenance costs.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
p-ISSN: 2395-0072
Volume: 09 Issue: 05 | May2022 www.irjet.net
PRAMOD MOHANTY, SRUJAN MHASE, SHANTANU PATIL, SAHIL YELGONDA (Students of DBIT)
Under Guidance of- Prof. JITHIN ISAAC, Dept. of EXTC,
Don Bosco Institute of Technology, Mumbai, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The motivation behind our task comes from the
innately unquenching prerequisite of expanding functional
productivity of a modern establishment. Notwithstanding
consistently expanding difficulties of building tough stock
chains that are getting more intricate constantly, ensuring
that the margin time of basic modern gear is grinding away's
base is perhaps the most noteworthy need of any
administrator. Remembering these difficulties, we aim to
develop a generalized, end-to-end plug and play system to
predict an outage before it happens and alert the operator of
any potential equipment malfunction using machine learning
algorithms deployed at edge.
Key Words: Vibration analysis, Embedded TinyML,
Condition monitoring, Maintenance, Arduino nano 33
BLE sense, Android app.
1.INTRODUCTION
Vibration analysis is a procedure that involves
detecting the vibration levels and frequencies of
machinery and then analysing how healthy the
machines and their components are. While the inner
workings and formulae used to calculate various types
of vibration might become intricate, it all begins with
the usage of an accelerometer to measure vibration by
adopting the method of Predictive Maintenance(PdM),
a process which involves continuously monitoring the
state of machinery to predict which parts will fail and
when. Maintenance can be planned in this way, and
only the parts that are showing signs of degradation or
damage can be replaced.
Predictive maintenance is based on taking
measurements that allow for the prediction of which
parts willfail and whenthey will fail.Machinevibration
and plant operational data such as flow, temperature,
and pressure are examples of these metrics.
The main benefits of PdM are: [4]
• Improved machine reliability through the effective
prediction of equipment failures [4]
•Reducedmaintenance costs byminimisingdowntime
through the scheduling of repairs [4]
• Increased production through greater machine
availability [4]
• Lower energy consumption [4]
• Improved product quality [4]
1.1 Motivation
Both IoT and Machine Learning are bringing about a
paradigm shift in the landscape. Getting hands on experience
with these technologies and deploying them will enhance
academic and career prospects.
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1709
Whenever a piece of equipment is in activity, it
produces vibrations. An accelerometer connected to
themachinedelivers a voltage signal that relates tothe
amount of vibration and therecurrenceofvibrationthe
machine produces, which is generally the times each
second or moment the vibration occurs. [4] The
accelerometer's data is taken care of promptly into an
information authority (programming), which catches
the sign as amplitude versus time (time waveform),
frequency versus recurrence (Fast Fourier change), or
both. All of this data is handled by PC modified
calculations, which are then checked on by engineers
or talented vibration experts to decide the machine's
wellbeing and recognize potential faults, for example,
detachment, lopsidedness, misalignment, grease
concerns, and more. Vibration analysers may now
accumulate, investigate, and convey information
substantially more effectively on account of
advancements in innovation, outstandingly remote
innovation. Vibration analysers are presently
amazingly convenient, can associateprogressivelywith
cell phones and tablets, and can make incredibly high-
goal FFT[4]. Numerous vibrations instrument
producers make their own applications to associate
with each other. Most of vibration examination
information is quickly shipped off the cloud and is
accessible on your cell phone, PC, or straightforwardly
from your program, just like with most trend setting
innovations. Assuming that you're performing
vibration investigation as an outsider expert, this is
very convenient in light of the fact that you may
uninhibitedly share spectra with your clients. [4]
Vibration Analysis for condition Monitoring & Predictive Maintenance
using Embedded TinyML](https://image.slidesharecdn.com/irjet-v9i5445-221005090628-fb16c9a9/85/Vibration-Analysis-for-condition-Monitoring-Predictive-Maintenance-using-Embedded-TinyML-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
p-ISSN: 2395-0072
Volume: 09 Issue: 05 | May2022 www.irjet.net
Fig -3: Node-Red website dashboard
Fig -4: Android application
ACKNOWLEDGEMENT
A project is a collaboration which incorporates the
responsibility of numerous people. We need to thank each
person who have contributed by looking at our work and
prodding us the entire way through. Ouractual appreciation
to our project guide Prof. Jithin Saji Isaac, for pushing,
collaborating and coordinating all of us through the errand
work, with his fruitful capacities and enormous data base.
We additionally need to express our gratitude to Dr.Ashwini
Kotrashetti, Head of Department, and Ms.Freda Carvalho,
project Coordinator for their consistent critical bearing,
sponsorship, thoughts and their significant time all over all
through the errand activity.
REFERENCES
[1] NarimanL. Dehghaniand Yousef Mohammadi
labels is first put away in the information base, making a
new single record with live sensor information and
expectations for each reading. Second This sensor is
additionally refreshed on a solitary JSON record which is
expected to show the live sensor readings and comparing
forecasts on the android application. The android
application is created with the assistance of the flutter app
development tool and the dart programming language.
The Android app displays real-time sensor data as well as
anomaly forecasts. Because the sensors are assigned to
separate stages, the user can now check for discrepancies
in sensor data in the Node Red interface. To view the
sensor’s live data, simply enter the proper tag. This can
be accomplished with any number of sensors.
3. CONCLUSIONS
Predictive maintenance and condition monitoring of heavy
machinery are clearly needed in enterprises. Manufacturers
are continually striving to increase output by maximizing
resource utilization. Companies can save time and costs by
keeping equipment downtime to a minimum, allowing them
to service more consumers. TinyML-based technologies let
us to carry out sophisticated tasks on low-powered portable
devices, saving us money on expensive systems and
equipment.
Keeping this in mind, we’ve briefly discussed the necessity
for predictive maintenance and our project, followed by a
thorough examination of existing solutions. We then
considered various solutions and offered a structure for
implementing our idea from the standpoint of product
development.
Darestani
and Abdollah Shafieezadeh Optimal Life-Cycle Resilience
Enhancement of Aging Power Distribution Systems: A
MINLP-Based Preventive Maintenance Planning Institute of
Electrical and Electronics Engineers (IEEE), 8th Volume,
2020 10.1109/access.2020.2969997.
[2] Pablo Aqueveque and Luciano Radrigan and Francisco
Pastene and Anibal S. Morales and Ernesto Guerra Data
Driven Condition Monitoring of Mining Mobile Machinery in
Non-Stationary Operations Using Wireless Accelerometer
Sensor Modules Institute of Electrical and Electronics
Engineers (IEEE), 9th Volume, 2021
10.1109/access.2021.3051583
[3] Ivar Koene and Ville Klar and Raine Viitala, IoT
connected device for vibration analysis and measurement
Elsevier BV, Volume 7, 2020, 10.1016/j.ohx 2020.e00109.
[4] Martin Pech and Jaroslav Vrchota and Jiˇr´ı Bedn´aˇr
Predictive Maintenance and Intelligent Sensors in Smart
Factory: Review MDPI AG, Sensors, Volume 21, Number 4,
2021, 10.3390/s21041470.
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1711](https://image.slidesharecdn.com/irjet-v9i5445-221005090628-fb16c9a9/85/Vibration-Analysis-for-condition-Monitoring-Predictive-Maintenance-using-Embedded-TinyML-3-320.jpg)
Vibration Analysis for condition Monitoring & Predictive Maintenance using Embedded TinyML
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 p-ISSN: 2395-0072 Volume: 09 Issue: 05 | May2022 www.irjet.net PRAMOD MOHANTY, SRUJAN MHASE, SHANTANU PATIL, SAHIL YELGONDA (Students of DBIT) Under Guidance of- Prof. JITHIN ISAAC, Dept. of EXTC, Don Bosco Institute of Technology, Mumbai, Maharashtra, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - The motivation behind our task comes from the innately unquenching prerequisite of expanding functional productivity of a modern establishment. Notwithstanding consistently expanding difficulties of building tough stock chains that are getting more intricate constantly, ensuring that the margin time of basic modern gear is grinding away's base is perhaps the most noteworthy need of any administrator. Remembering these difficulties, we aim to develop a generalized, end-to-end plug and play system to predict an outage before it happens and alert the operator of any potential equipment malfunction using machine learning algorithms deployed at edge. Key Words: Vibration analysis, Embedded TinyML, Condition monitoring, Maintenance, Arduino nano 33 BLE sense, Android app. 1.INTRODUCTION Vibration analysis is a procedure that involves detecting the vibration levels and frequencies of machinery and then analysing how healthy the machines and their components are. While the inner workings and formulae used to calculate various types of vibration might become intricate, it all begins with the usage of an accelerometer to measure vibration by adopting the method of Predictive Maintenance(PdM), a process which involves continuously monitoring the state of machinery to predict which parts will fail and when. Maintenance can be planned in this way, and only the parts that are showing signs of degradation or damage can be replaced. Predictive maintenance is based on taking measurements that allow for the prediction of which parts willfail and whenthey will fail.Machinevibration and plant operational data such as flow, temperature, and pressure are examples of these metrics. The main benefits of PdM are: [4] • Improved machine reliability through the effective prediction of equipment failures [4] •Reducedmaintenance costs byminimisingdowntime through the scheduling of repairs [4] • Increased production through greater machine availability [4] • Lower energy consumption [4] • Improved product quality [4] 1.1 Motivation Both IoT and Machine Learning are bringing about a paradigm shift in the landscape. Getting hands on experience with these technologies and deploying them will enhance academic and career prospects. © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1709 Whenever a piece of equipment is in activity, it produces vibrations. An accelerometer connected to themachinedelivers a voltage signal that relates tothe amount of vibration and therecurrenceofvibrationthe machine produces, which is generally the times each second or moment the vibration occurs. [4] The accelerometer's data is taken care of promptly into an information authority (programming), which catches the sign as amplitude versus time (time waveform), frequency versus recurrence (Fast Fourier change), or both. All of this data is handled by PC modified calculations, which are then checked on by engineers or talented vibration experts to decide the machine's wellbeing and recognize potential faults, for example, detachment, lopsidedness, misalignment, grease concerns, and more. Vibration analysers may now accumulate, investigate, and convey information substantially more effectively on account of advancements in innovation, outstandingly remote innovation. Vibration analysers are presently amazingly convenient, can associateprogressivelywith cell phones and tablets, and can make incredibly high- goal FFT[4]. Numerous vibrations instrument producers make their own applications to associate with each other. Most of vibration examination information is quickly shipped off the cloud and is accessible on your cell phone, PC, or straightforwardly from your program, just like with most trend setting innovations. Assuming that you're performing vibration investigation as an outsider expert, this is very convenient in light of the fact that you may uninhibitedly share spectra with your clients. [4] Vibration Analysis for condition Monitoring & Predictive Maintenance using Embedded TinyML
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 p-ISSN: 2395-0072 Volume: 09 Issue: 05 | May2022 www.irjet.net Increasing amounts arebeing spentonimprovingoperational efficiency of industrial set-ups. Providing cutting edge predictivemaintenance solutions may help uspositionthisas a product. 1.2 Objective and Outcome Objective: Build an Anomaly Detection Model using sklearn on Google Colab Export sklearn model and run inference for prototyping and debugging onboard the microcontroller-Arduino nano 33 ble sense Mobile Desktop Interface for condition monitoring environment setting for users Implementing CI/CD pipeline for periodic data collection and training for improving model accuracy Outcome: an end-to-end condition monitoring and predictive maintenance system using machine learning for industrial equipment with a mobile & desktop application for user interface giving access of condition monitoring for numerous equipment at once where system performance increases with time as more and more data is collected. 2. METHODOLOGY The test bench is used to guide the project’s operations.Two scenarios are used to obtain data from the test bench. The servo motor is turned off at first, and this is the typical working state of the DC motor. When the servo motor is turned on, it begins to shake the board, causing unnatural vibrations. Fig -1: Block diagram The accelerometer captures this vibration data using the Arduino nano 33 BLE sense. The data is then saved and converted to a csv file in order to train the ML model. The CSV file is read after it has been loaded into the Python function. The data is then utilized to train a machinelearning model using Python’s sklearn module and the randomforest classifier. Once the model has been trained the Micromlgen’s Python library is used to transform the ML model into an Arduino header fileafter it has been developed. TheArduino code then uses this header file to compare the data from the live sensor (accelerometer). Based on the output, itforecasts either an anomaly 1 or no anomaly 0. Chart -1: X, Y, Z -axis accelerometer reading with normal state and servo on. Fig -3: Test bench setup © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1710 The live sensor information and the predictions are then changed into JSON string and sent to the serial monitor. Whenever Node Red is working on the local host port 1880, information from the sensor is gathered on the sequential port. Since the information is in a string format, a node is utilized to change it over completely to a JSON string and transform it into a JSON record. This data is used to refresh the site with ongoing information that clients might see alongside expectations and accelerometer information on three axes. This data is likewise given to the Firebase ongoing information administration. The firebase stores this data in two variants. First for future reference, a JSON report with randomly generated
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 p-ISSN: 2395-0072 Volume: 09 Issue: 05 | May2022 www.irjet.net Fig -3: Node-Red website dashboard Fig -4: Android application ACKNOWLEDGEMENT A project is a collaboration which incorporates the responsibility of numerous people. We need to thank each person who have contributed by looking at our work and prodding us the entire way through. Ouractual appreciation to our project guide Prof. Jithin Saji Isaac, for pushing, collaborating and coordinating all of us through the errand work, with his fruitful capacities and enormous data base. We additionally need to express our gratitude to Dr.Ashwini Kotrashetti, Head of Department, and Ms.Freda Carvalho, project Coordinator for their consistent critical bearing, sponsorship, thoughts and their significant time all over all through the errand activity. REFERENCES [1] NarimanL. Dehghaniand Yousef Mohammadi labels is first put away in the information base, making a new single record with live sensor information and expectations for each reading. Second This sensor is additionally refreshed on a solitary JSON record which is expected to show the live sensor readings and comparing forecasts on the android application. The android application is created with the assistance of the flutter app development tool and the dart programming language. The Android app displays real-time sensor data as well as anomaly forecasts. Because the sensors are assigned to separate stages, the user can now check for discrepancies in sensor data in the Node Red interface. To view the sensor’s live data, simply enter the proper tag. This can be accomplished with any number of sensors. 3. CONCLUSIONS Predictive maintenance and condition monitoring of heavy machinery are clearly needed in enterprises. Manufacturers are continually striving to increase output by maximizing resource utilization. Companies can save time and costs by keeping equipment downtime to a minimum, allowing them to service more consumers. TinyML-based technologies let us to carry out sophisticated tasks on low-powered portable devices, saving us money on expensive systems and equipment. Keeping this in mind, we’ve briefly discussed the necessity for predictive maintenance and our project, followed by a thorough examination of existing solutions. We then considered various solutions and offered a structure for implementing our idea from the standpoint of product development. Darestani and Abdollah Shafieezadeh Optimal Life-Cycle Resilience Enhancement of Aging Power Distribution Systems: A MINLP-Based Preventive Maintenance Planning Institute of Electrical and Electronics Engineers (IEEE), 8th Volume, 2020 10.1109/access.2020.2969997. [2] Pablo Aqueveque and Luciano Radrigan and Francisco Pastene and Anibal S. Morales and Ernesto Guerra Data Driven Condition Monitoring of Mining Mobile Machinery in Non-Stationary Operations Using Wireless Accelerometer Sensor Modules Institute of Electrical and Electronics Engineers (IEEE), 9th Volume, 2021 10.1109/access.2021.3051583 [3] Ivar Koene and Ville Klar and Raine Viitala, IoT connected device for vibration analysis and measurement Elsevier BV, Volume 7, 2020, 10.1016/j.ohx 2020.e00109. [4] Martin Pech and Jaroslav Vrchota and Jiˇr´ı Bedn´aˇr Predictive Maintenance and Intelligent Sensors in Smart Factory: Review MDPI AG, Sensors, Volume 21, Number 4, 2021, 10.3390/s21041470. © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1711