11 IV April 2023

https://doi.org/10.22214/ijraset.2023.51254
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue IV Apr 2023- Available at www.ijraset.com

Predictive Machine Maintenance Using Tiny ML

Abhishek Bhide1, Dnyaneshvar Ghodake2, Ashish Jamle3, Salman Shaikh4, Prof.S.R.Bhujbal5
1, 2, 3, 4
Students, B.E. Computer, 5Professor, Dept. of Computer Engineering, P.K Technical Campus Pune, Maharashtra

Abstract: Anomaly detection (AD) is detection of pattern in data in expected behaviour. In an industrial environment, any
equipment and system that breaks down are affecting productivity. Therefore, Tiny Machine Learning (Tiny ML) is introduced
to address this problem. Tiny ML undergo anomaly detection to detect if any equipment did not act expected behaviour and
notify the user if an anomaly detection has been detected. Anomaly detection is an unsupervised learning algorithm. It has aim
to identify the patterns of data that do not follow the expected behaviour.
Tiny Machine Learning (Tiny), a rapidly evolving edge computing concept that links embedded systems (hardware and
software) and machine learning, with purpose of realizing ultra-low-power or low-cost and efficiency and privacy also brings
machine learning inference to battery-powered intelligent devices.
By using TensorFlow Lite Micro, the Tiny ML can be trained to undergo anomaly detection. How is the machine learning
process has exported for TensorFlow, then TensorFlow and final TensorFlow L Micro order to upload the machine learning
algorithm with Tiny ML.
The paper highlight the state the art of the current work on Tiny Machine Learning.
Keywords: ESP32, Accelerometer, Temperature, Buzzer, LED, etc.

I. INTRODUCTION
Reliability, Which is the main attribute of safe operation in any modern technological system. Reliability analysis focus on
incertainty in failure incidences and the aftereffects. The aim shield the systems beyond the incertainties of there accidental
situations. In recent years, reliability engineering has been well established as a multi-disciplinary scientific discipline that seeks to
offer an ensemble of formal methods to inspect the unclear boundaries between system operation and failure. Furthermore, E. Zio is
also recorded to main question focusing in reliability engineering.
These questions are the main causesof the system’s failure, the procedure of measuring and testing the reliability in design, operation,
and management, the ways to maintain the system’s reliability such as maintenance, fault diagnosis, and prognosis, and the
techniques to develop the reliable systems. Engineering maintenance is prognostics plays a critical role of modern industries such as
aerospace, loco motive, manufacture, and so etc. The usually engineering maintenance approach is the maintain of functionality the
equipment or system, such as preventive maintenance.
Tiny machine learning is rapidly growing field machine learning technologies and applications that include hardware , algorithms,
and software capable of performing on-device sensor (camera/colour vision, audio, IMU, biomedical) data analytics of extreme low
power, typically the mw range as below, enable the variety of a always-on use-cases of targeting battery-powered devices.It is
delivering low-latency, low-powers, and low-bandwidths model inference at edge devices.
The typicaly microcontroller consumes electricity the milliwatts the microwatts range, whereas a typical consumer CPU consumes
in range 65 and 85 watts and a typical consumer GPU consumes between 200 to 500 watts. This equates a thousand-fold reduction
in power use.

II. METHODOLOGY
In this proposed system we have proposed a new concept through tiny ML, in which we are implementing Predictive Machine
Maintenance in which an ESP-32 Microcontroller is used to control the flow of the Accelerometer.
Through this, we can read Accelerometer Data and Then Tensor flow lite Interpreter which runs the model. Then it classifies the
model which is Trained to Normal or Abnormal Activity.
Whenever the model is in the normal state then Interpreter sends the message to the Model and glows Green Led Light and Buzzer
off, and if the model is in an abnormal state, then the model shows Red LED and Buzzers.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 4252
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue IV Apr 2023- Available at www.ijraset.com

Fig 1. Data Flow Diagram

III. PROPOSED SYSTEM

To build predictive machine maintenance system using tiny ML. In this system we are predict machine maintenance by a sensing
vibration of machine. we are using accelerometer sensor to collect data , then build and optimize ML model to deploy to edge
hardware.

A. ESP32
ESP32 would be a series of low-cost, low-power systems with on-chip microcontrollers in integrated Wi-Fi or dual-modeBluetooth.

B. Piezo Buzzer
The Grove Buzzer module has piezo buzzer as it is main component. The piezo element makes a clicking sound every time when it
is pulsed with current. If we pulse it at the correct frequency, these frequent clicks will run together to develop a little melody.

C. Accelerometer
An accelerometer sensor is the tool that measure the acceleration of anybody or an object in its instantaneous rest frame. It is not a
coordinate acceleration. Accelerometer sensor is used in many electronic devices, smartphones and wearable devices, etc.
Accelerometers is used in biomedical applications, biomedical field accelerometer sensors is mainly operated in step counting or
activity monitoring.

D. Male Female Wire

This cable is an electrical device or wire and group of the cable with a connector and also with pins at every last stage, which is
normal for interconnecting t componentsof a breadboard and other prototype and test circuit, internal with other equipment are
components which is without soldering.

E. LED
LED adjust with brightness as well as adjustable orientation. Then, Grove LED is perfect for your need. The Grove and Red LED
are designed for monitoring controls from digital ports. It is very helpful for beginners of Arduino and Seeding. The PCB of this
module is mounting hole, so can you mount it with the required surface on a box or the desk and a prototype.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 4253
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue IV Apr 2023- Available at www.ijraset.com

Accelerometer Sensor Read Accelerometer data

TF Lite Interpreter Model Trained to

Runs the model Classify Normal ,
abnormal Activity

Identify normal/abnormal
activity

LED Buzzer Play Command Responder

Fig.2 System architecture

In this proposed system we have proposed a new concept through tiny ML, in which we are implementing Predictive Machine
Maintenance in which ESP-32 Microcontroller is used to control the flow of Accelerometer. Through which we can read
Accelerometer Data then Tensor flow lite Interpreter which runs the model. Then it classifies the model which is Trained to Normal
orAbnormal Activity. Whenever the model is in normal state then Interpreter sends the message to Model and glows Green Led.

IV. RESULTS
Implementation of tiny machine learning era in predictive platform growth contributor’s confidentiality whilst contributing to
machine learning. This is because of the character of tiny ML is easy. Predictive Maintenance techniques detect anomalies in
equipment first which are turn in the system-critical failures, allowing maintenances which can be schedule first the equipment
actually break down. The increase equipment uptime do less of all the maintenance costs and allow optimization of the part inventor
by enablepreventivemachine maintenance based on the equipment’s which needed. After having prove it has benefit in context of
Industry. the data-driven maintenance strategy is adopted in the Smart Buildings of tomorrow. Especially in infrastructure critical
assets which has HVAC, Predictive machine Maintenance offers high benefits. Condition monitoring use the sensors which is to
provide meaningful insights in the current health of different devices and items of equipment of the buildings. The sensor is
collecting the data of monitor crucial operating parameters which as vibrations, sound anomalies, airflow, and current. The various
parts of predictive machine maintenance takes condition-based maintenanceto the nextlevel.

V. CONCLUSIONS
We presented a deep learning-based approach for the detection of anomalies in time series data. Since the approach is
unsupervised, it requires no labels for anomalies. Instead, the method models the regular data distribution and marks data points
which don’t conform to this model as anomalous.

REFERENCES
[1] Hamdan, S.; Ayyash, M.; AL Majali, S. Edge-Computing Architectures for Internet of Things Applications: A Survey. Sensors 2020, 20, 6441.
[2] Wu, Z.; Qiu, K.; Zhang, J. A Smart Microcontroller Architecture for the Internet of Things. Sensors 2020, 20, 1821.
[3] Signoretti, G.; Silva, M.; Andrade, P.; Silva, I.; Sisinni, E.; Ferrari, P. An Evolving Tiny ml Compression Algorithm for IoT Environments Based on Data
Eccentricity. Sensors 2021, 21, 4153.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 4254
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 11 Issue IV Apr 2023- Available at www.ijraset.com

[4] Chen, Y.; Zheng, B.; Zhang, Z.; Wang, Q.; Shen, C.; Zhang, Q. Deep Learning on Mobile and Embedded Devices: State-of-the-Art, Challenges, and Future
Directions. ACM Comput. Surv. 2020, 53, 1–37.
[5] N. R. Prasad, S. Almanza-Garcia, and T. T. Lu, “Anomaly detection,” Comput. Mater. vol. 14, no. 1, pp. 1–22, 2009, doi: 10.1145/1541880.1541882.
[6] S. Soro, “TinyML for Ubiquitous Edge AI,” no. 20, 2021, [Online].
[7] R. Sanchez-Iborra and A. F. Skarmeta, “TinyML-Enabled Frugal Smart Objects: Challenges and Opportunities,” IEEE Circuits and Systems Magazine, vol. 20,
2020.
[8] Y. Zhang, N. Suda, L. Lai, and V. Chandra, “Hello Edge: Keyword Spotting on Microcontrollers,” 2018. [Online]. Available: arXiv:1711.07128
[9] G. Selander, J. Mattsson, F. Palombini, and L. Seitz, “Object security for constrained restful environments (score),” Work in Progress, 2019
[10] Sanchez-Iborra and A. F. Skarmeta, “TinyML-Enabled Frugal Smart Objects: Challenges and Opportunities,” IEEE Circuits and Systems Magazine, vol. 20,
2020
[11] M. S. Mahdavinejad, M. Rezvan, M. Barekatain, P. Adibi, P. Barnaghi, and A. P. Sheth, “Machine learning for internet of things data analysis: a survey,”
Digital Communications and Networks, vol. 4, 2018.
[12] L. Dutta and S. Bharali, ‘‘Tiny meets IoT: A comprehensive survey,’’ 705 Internet Things, vol. 16, Dec. 2021, Art. no. 100461.
[13] M. Tschannen, A. Khanna, and A. Anandkumar, ‘‘Strassen Nets: Deep 683 learning with multiplication budget,’’ in Proc. 35th Int. Conf. Mach. 684 Learn.,
Stockholm, Sweden, 2018, pp. 4985–4994.
[14] T. F. Foundation. Tflite Micro Implementation Tutorials. Accessed: 719 Aug. 23, 2022.
[15] X. Wang, M. Magno, L. Cavigelli, and L. Benini, ‘‘FANN-on-MCU: An 650 open-source tool kit for energy-efficient neural network inference at the 651 edge
of the Internetofhings,’’ IEEE Internet Things..
[16] Vu, T.H.; Olsson, C.; Laptev, I.; Oliva, A.; Sivic, J. Food-101–Mining Discriminative Components with Random Forests; LNCS 8693; Springer International
Publishing: Zurich, Switzerland, 2014.
[17] Kievsky, A. Learning Multiple Layers of Features from Tiny Images; University of Toronto: Toronto, ON, Canada, 2009;Volume 34.
[18] Chowdhary, A.; Warden, P.; Sheens, J.; Howard, A.; Rhodes, R. Visual Wake Words Dataset. aria 2019,arXiv:1906.05721v1.
[19] Boero, L.; Cello, M.; Marchese, M.; Marcionite, E.; Nevash, T.; Zappa tore, S. Statistical Fingerprint-Based Intrusion Detection System (SF-IDS). Int.
J.Common. Syst.2016,30,e3225.
[20] health.; Klisch, D.; Kubik, P.; Ali, A.; Messier, J.D.; Turau, V. Artificial Neural Networks for Sensor Data Classification on Small Embedded Systems. arci
2020, arXiv:2012.08403v1

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 4255