SlideShare ist ein Scribd-Unternehmen logo
1 von 27
Downloaden Sie, um offline zu lesen
Smart Manufacturing 4.0
RISHABH GARG
BITS - PILANI I GOA
IIT Delhi - AIA
Fault Detection of
Smart Lathe Machine
Using Python
Rishabh Garg
BITS - PILANI
K.K. Birla Goa Campus
IIT Delhi - AIA Foundation For Smart Manufacturing
.………………
Best Project Award
Samarth Udyog 4.0
Indian Institute of
Technology Delhi
Department of Heavy Industry
Ministry of HI & PE
Automation Industry
Association
A Government of India
Initiative Hosting Institute Government of India Industry Partner
1
3
A
U
G
U
S
T
2
0
2
1
Fault Detection of Smart Lathe
Machine using Python
Rishabh Garg
BITS - PILANI | GOA
IITD-AIA Foundation For Smart Manufacturing
Challenge
RISHABH GARG | BITS
Project Background
Problem Statement
 Machines having large investments in capital also
deploy high skilled labor for active operation and
periodic maintenance. The challenge for companies is to
protect this investment by ensuring high availability.
 Unpreceded breakdowns in machinery and lack of
skilled labor becomes a major challenge for
manufacturing industries. Customer orders get delayed
and company loses money
 Solution: Deploy Machine Learning models and Real
time dashboards and Predictive maintenance that
constantly monitor the health of the machine and the
associated tools.
RISHABH GARG | BITS
Project Background
Physical Equipment on which it is implemented
 No specific hardware required, apart from a
regular PC.
 GPUs can be used for faster training and results.
 OS: Ubuntu – 18.04 LTS and Google
Collaboratory (with GPUs)
Unique challenge
Development of accurate model despite high bias and
limited dataset.
RISHABH GARG | BITS
LOREM IPSUM DOLOR SIT AMET,
CONSECTETUER ADIPISCING ELIT. 12
Lorem ipsum dolor sit amet, consectetuer adipiscing elit.
Maecenas porttitor congue massa. Fusce posuere, magna
sed pulvinar ultricies, purus lectus malesuada libero, sit
amet commodo magna eros quis urna.
7
Solution
RISHABH GARG | BITS
Objective
 Predict the instantaneous RMS value of vibrations
of a Smart Lathe Machine using the traditional
input parameters
 Use the prediction for classification of the
machine state into Normal and Faulty behavior.
Objective
 Display the prediction results through Streamlit
Web App.
 Implement the API endpoints using Flask.
 Deployment to Heroku using Docker containers.
RISHABH GARG | BITS
Methodology
Exploration of data
 Data extraction from CSV files, Data preparation
and filtering
 Representation of data and its analysis (kurtosis,
skewness etc. for appropriate model selection)
Machine Learning Model Development
Developed some sample machine learning models for
checking accuracy and other metrics like sensitivity
or positivity rate. Research was done on the models
for hyper-parameter tuning etc.
RISHABH GARG | BITS
Methodology
Flask
Model was packed and API implementation was done
using Flask
Deployment through Docker containers
Streamlit
Necessary pathways for connection with dashboard
using Streamlit
Deployment to Heroku was done using CLI
RISHABH GARG | BITS
Implementation
Data Extraction
 Dataset provided from the FSM Lab consisted of
raw data which was directly taken from the DAC
device connected to the accelerometer.
 For machine learning models to be built, first the
required input parameters were identified.
 RMS value of vibration was the best output
parameter for predicting the machine state.
RISHABH GARG | BITS
Implementation
Preparation of raw data
 Using Python OS library, the vibration data stored in
CSV files was aggregated and summarized in an Excel
file.
 60 One-Minute Experiments used as the training data for
the linear regression model.
 Redundant columns such as Dates, Timestamps, Serial
numbers, Status etc. were removed and only numeric
columns were retained.
 RMS buffer values and Actual RMS values were used as
output variables.
RISHABH GARG | BITS
Implementation
Preparation of raw data
 One-Minute experiments were used to predict the
initial level of vibrations.
 Full run experiments were used to determine the
slope of RMS values with time.
 RMS value were compared against the value
provided by the line at a given time for
classification.
RISHABH GARG | BITS
Implementation
Preparation of raw data
 Note: The data was very biased due to low occurrences
of failure in the Lathe machine. This affected the
accuracy of the model in the later stages.
RISHABH GARG | BITS
Implementation
Training the model
 Cubic SVM and Linear Regression models were
compared for performance using MAE (Mean Absolute
Error), RMSE (Root Mean Squared Error) metrics.
 Linear Regression model was chosen out of the two due
to higher dimensionality of data.
 The threshold value of RMS vibration came out to be
around 372.86 for the classifier model.
 Logistic Regression was used to implement the classifier
layer of model.
 The predicted / actual RMS values were fed into the
classifier for prediction of the machine state.
RISHABH GARG | BITS
Implementation
Testing the model
 After training both the models, the test dataset, generated
using
 Scikit learn library was fed the model and the predictions
were stored.
 The models were also
evaluated based on a few
metrics from the scikit
learn library like MAE,
R2, RMSE (for
regression) and F1 score
(for classification).
RISHABH GARG | BITS
Implementation
Deployment
 Heroku was used for the deployment of the final
deliverable product which was connected to a Git
repository for storage of the codebase.
 A Docker file, Procfile and requirements.txt file
was required for the appropriate stack selection by
Heroku.
 Streamlit was used for creation of the deployed
frontend webapp and Flask was used to make the
required API calls.
RISHABH GARG | BITS
Implementation
Deployment
RISHABH GARG | BITS
20
Attainments
RISHABH GARG | BITS
Innovations
 Real time monitoring of the input parameters,
instantaneous and accurate prediction of RMS values
of vibration data along with display of associated
metrics was the biggest industrial level achievement of
this product.
 One can also
select multiple
models for
prediction in
the web app
along with
intuitive
visualization.
RISHABH GARG | BITS
Innovations
 This model, despite the associated bias, provides
decent predictions due to proper choice of degree of
SVR (here, behaves like Linear regressor). Tunable
parameters have been added as sidebars in the webapp
for customized accuracy as per the use case.
RISHABH GARG | BITS
Outcome
Final deliverable product
The following applications and repositories were made
on the completion of the internship project for
addressing the problem statement:
 The app was deployed to Heroku as a frontend
webapp: https://streamlitappml.herokuapp.com/
 Source code for the Base models, Docker files and
Flask: https://github.com/rishabhgargdps/IITD_ML
RISHABH GARG | BITS
Outcome
Final deliverable product
Source code for the Front end application:
https://github.com/rishabhgargdps/IITD_streamlit
Heroku app for customized models:
https://streamlitml.herokuapp.com/
Heroku flask app that contains the API, and Docker
containers: https://svmml2.herokuapp.com/
The codebase of the project was stored in Git Version
control systems (here, GitHub).
RISHABH GARG | BITS
Scalability
 This model can be used for prediction of RUL of any
modern / traditional machine / tools if we can
identify the most relevant input and output
parameters pertaining to the quality of the product
and the life of the machine.
 The vibration data can be collected from the
accelerometer attached to the tool holder which can
be sent to the cloud for data storage using the
Ethernet. There it can be processed and used for
automated prediction.
 Additional parameters like temperature, humidity of
environment can be added for even better
predictions. Thermo-hygrometers (DHT-11) can be
used in the rooms for temperature and humidity
monitoring.
RISHABH GARG | BITS
Scalability
 One can also adjust the model for false positives /
negatives as per need in the case of legacy
hardware / biological equipment manufacturing.
 The operator can be alerted using a web monitor
service (e.g. Distil web monitor) running on the
web app through an alarm sound, email
notification etc.
 Pre-trained neural nets or decision trees in a
random forest can be used if we cannot acquire a
large dataset and a similar use case has already
been solved for a large dataset.
RISHABH GARG | BITS
Thank you !

Weitere ähnliche Inhalte

Was ist angesagt?

Transformer protection using microcontroller and gsm technology
Transformer protection using microcontroller and gsm technologyTransformer protection using microcontroller and gsm technology
Transformer protection using microcontroller and gsm technology
Kartik Patel
 
Performance analysis and prediction of stock market for investment decision u...
Performance analysis and prediction of stock market for investment decision u...Performance analysis and prediction of stock market for investment decision u...
Performance analysis and prediction of stock market for investment decision u...
Hari KC
 
Credit card fraud detection
Credit card fraud detectionCredit card fraud detection
Credit card fraud detection
kalpesh1908
 

Was ist angesagt? (20)

Power System Modelling And Simulation Lab
Power System Modelling And Simulation LabPower System Modelling And Simulation Lab
Power System Modelling And Simulation Lab
 
FOUR QUADRANT DC MOTOR SPEED CONTROL WITH MICROCONTROLLER
FOUR QUADRANT DC MOTOR SPEED CONTROL WITH MICROCONTROLLERFOUR QUADRANT DC MOTOR SPEED CONTROL WITH MICROCONTROLLER
FOUR QUADRANT DC MOTOR SPEED CONTROL WITH MICROCONTROLLER
 
CSI and vSI fed induction motor drives
CSI and vSI fed induction motor drivesCSI and vSI fed induction motor drives
CSI and vSI fed induction motor drives
 
Selective Harmonic Elimination Based on Newton-raphson Method for Cascaded H-...
Selective Harmonic Elimination Based on Newton-raphson Method for Cascaded H-...Selective Harmonic Elimination Based on Newton-raphson Method for Cascaded H-...
Selective Harmonic Elimination Based on Newton-raphson Method for Cascaded H-...
 
Coin based mobile charger project report
Coin based mobile charger project reportCoin based mobile charger project report
Coin based mobile charger project report
 
Data Mining
Data MiningData Mining
Data Mining
 
Electric locomotive ppt
Electric locomotive pptElectric locomotive ppt
Electric locomotive ppt
 
Transformer protection using microcontroller and gsm technology
Transformer protection using microcontroller and gsm technologyTransformer protection using microcontroller and gsm technology
Transformer protection using microcontroller and gsm technology
 
Market Basket Analysis
Market Basket AnalysisMarket Basket Analysis
Market Basket Analysis
 
Performance analysis and prediction of stock market for investment decision u...
Performance analysis and prediction of stock market for investment decision u...Performance analysis and prediction of stock market for investment decision u...
Performance analysis and prediction of stock market for investment decision u...
 
Big data unit i
Big data unit iBig data unit i
Big data unit i
 
Credit card fraud detection
Credit card fraud detectionCredit card fraud detection
Credit card fraud detection
 
Matlab for Electrical Engineers
Matlab for Electrical EngineersMatlab for Electrical Engineers
Matlab for Electrical Engineers
 
Electric traction doc/sanjeet-1308143
Electric traction doc/sanjeet-1308143Electric traction doc/sanjeet-1308143
Electric traction doc/sanjeet-1308143
 
20IT501_DWDM_PPT_Unit_V.ppt
20IT501_DWDM_PPT_Unit_V.ppt20IT501_DWDM_PPT_Unit_V.ppt
20IT501_DWDM_PPT_Unit_V.ppt
 
Data Mining
Data MiningData Mining
Data Mining
 
OPAL-RT Power Hardware-In-the-Loop presentation
OPAL-RT Power Hardware-In-the-Loop presentationOPAL-RT Power Hardware-In-the-Loop presentation
OPAL-RT Power Hardware-In-the-Loop presentation
 
Presentation1
Presentation1Presentation1
Presentation1
 
Water pump automation system
Water pump automation systemWater pump automation system
Water pump automation system
 
UNIT2-Data Mining.pdf
UNIT2-Data Mining.pdfUNIT2-Data Mining.pdf
UNIT2-Data Mining.pdf
 

Ähnlich wie Fault Detection using Python

November 2013 HUG: Real-time analytics with in-memory grid
November 2013 HUG: Real-time analytics with in-memory gridNovember 2013 HUG: Real-time analytics with in-memory grid
November 2013 HUG: Real-time analytics with in-memory grid
Yahoo Developer Network
 
Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
Amazon Web Services
 

Ähnlich wie Fault Detection using Python (20)

"The Suitcase" Project Cloud QTR meeting presentation @ Disney/ABC
"The Suitcase"  Project Cloud QTR meeting presentation @ Disney/ABC"The Suitcase"  Project Cloud QTR meeting presentation @ Disney/ABC
"The Suitcase" Project Cloud QTR meeting presentation @ Disney/ABC
 
Machine Intelligence for Design Automation
Machine Intelligence for Design AutomationMachine Intelligence for Design Automation
Machine Intelligence for Design Automation
 
Presentation
PresentationPresentation
Presentation
 
QCon 2018 | Gimel | PayPal's Analytic Platform
QCon 2018 | Gimel | PayPal's Analytic PlatformQCon 2018 | Gimel | PayPal's Analytic Platform
QCon 2018 | Gimel | PayPal's Analytic Platform
 
November 2013 HUG: Real-time analytics with in-memory grid
November 2013 HUG: Real-time analytics with in-memory gridNovember 2013 HUG: Real-time analytics with in-memory grid
November 2013 HUG: Real-time analytics with in-memory grid
 
Using AWS IoT for Industrial Applications - AWS Online Tech Talks
Using AWS IoT for Industrial Applications - AWS Online Tech TalksUsing AWS IoT for Industrial Applications - AWS Online Tech Talks
Using AWS IoT for Industrial Applications - AWS Online Tech Talks
 
Operationalizing Your Analysis with AWS IoT Analytics (IOT358-R1) - AWS re:In...
Operationalizing Your Analysis with AWS IoT Analytics (IOT358-R1) - AWS re:In...Operationalizing Your Analysis with AWS IoT Analytics (IOT358-R1) - AWS re:In...
Operationalizing Your Analysis with AWS IoT Analytics (IOT358-R1) - AWS re:In...
 
VTT RobotDay 5.9. Harri Kuusela: DIH² ja L4MS hankkeiden tuki yrityksille
VTT RobotDay 5.9. Harri Kuusela: DIH² ja L4MS hankkeiden tuki yrityksilleVTT RobotDay 5.9. Harri Kuusela: DIH² ja L4MS hankkeiden tuki yrityksille
VTT RobotDay 5.9. Harri Kuusela: DIH² ja L4MS hankkeiden tuki yrityksille
 
Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
 
Automated Low Level Requirements Testing for DO-178C
Automated Low Level Requirements Testing for DO-178CAutomated Low Level Requirements Testing for DO-178C
Automated Low Level Requirements Testing for DO-178C
 
Design and Development of Multipurpose Automatic Material Handling Robot with...
Design and Development of Multipurpose Automatic Material Handling Robot with...Design and Development of Multipurpose Automatic Material Handling Robot with...
Design and Development of Multipurpose Automatic Material Handling Robot with...
 
Preventative Maintenance of Robots in Automotive Industry
Preventative Maintenance of Robots in Automotive IndustryPreventative Maintenance of Robots in Automotive Industry
Preventative Maintenance of Robots in Automotive Industry
 
AI Expo - AI Revolution in Silicon Valley
AI Expo - AI Revolution in Silicon ValleyAI Expo - AI Revolution in Silicon Valley
AI Expo - AI Revolution in Silicon Valley
 
Predicting medical tests results using Driverless AI
Predicting medical tests results using Driverless AIPredicting medical tests results using Driverless AI
Predicting medical tests results using Driverless AI
 
IRJET- A Review- FPGA based Architectures for Image Capturing Consequently Pr...
IRJET- A Review- FPGA based Architectures for Image Capturing Consequently Pr...IRJET- A Review- FPGA based Architectures for Image Capturing Consequently Pr...
IRJET- A Review- FPGA based Architectures for Image Capturing Consequently Pr...
 
Doctor ZedGe @InsideTrack Rome #sitROME
Doctor ZedGe @InsideTrack Rome #sitROMEDoctor ZedGe @InsideTrack Rome #sitROME
Doctor ZedGe @InsideTrack Rome #sitROME
 
Engineering plant facilities 17 artificial intelligence algorithms & proc...
Engineering plant facilities 17 artificial intelligence algorithms & proc...Engineering plant facilities 17 artificial intelligence algorithms & proc...
Engineering plant facilities 17 artificial intelligence algorithms & proc...
 
Engineering plant facilities 17 artificial intelligence algorithms & proc...
Engineering plant facilities 17 artificial intelligence algorithms & proc...Engineering plant facilities 17 artificial intelligence algorithms & proc...
Engineering plant facilities 17 artificial intelligence algorithms & proc...
 
ABAP Test Cockpit in action with Doctor ZedGe and abap2xlsx
ABAP Test Cockpit in action with Doctor ZedGe and abap2xlsxABAP Test Cockpit in action with Doctor ZedGe and abap2xlsx
ABAP Test Cockpit in action with Doctor ZedGe and abap2xlsx
 
Virtual BenchLearning - I-BiDaaS - Industrial-Driven Big Data as a Self-Servi...
Virtual BenchLearning - I-BiDaaS - Industrial-Driven Big Data as a Self-Servi...Virtual BenchLearning - I-BiDaaS - Industrial-Driven Big Data as a Self-Servi...
Virtual BenchLearning - I-BiDaaS - Industrial-Driven Big Data as a Self-Servi...
 

Mehr von Rishabh Garg

Mehr von Rishabh Garg (14)

Stocks : Technical Analysis
Stocks : Technical AnalysisStocks : Technical Analysis
Stocks : Technical Analysis
 
International Conference | Artificial Intelligence & Machine Learning
International Conference | Artificial Intelligence & Machine LearningInternational Conference | Artificial Intelligence & Machine Learning
International Conference | Artificial Intelligence & Machine Learning
 
Biosensors
BiosensorsBiosensors
Biosensors
 
Python Library using impedance processing
Python Library using impedance processingPython Library using impedance processing
Python Library using impedance processing
 
Documentation and Deployment through Python Libraries
Documentation and Deployment through Python LibrariesDocumentation and Deployment through Python Libraries
Documentation and Deployment through Python Libraries
 
International Webinar - Global ID Through Blockchain
International Webinar - Global ID Through BlockchainInternational Webinar - Global ID Through Blockchain
International Webinar - Global ID Through Blockchain
 
International Talk on Technical Analysis
International Talk on Technical AnalysisInternational Talk on Technical Analysis
International Talk on Technical Analysis
 
Machine Learning Applications
Machine Learning ApplicationsMachine Learning Applications
Machine Learning Applications
 
NAAC : Assessment & Accreditation
NAAC : Assessment & AccreditationNAAC : Assessment & Accreditation
NAAC : Assessment & Accreditation
 
NAAC : Accreditation Process
NAAC : Accreditation ProcessNAAC : Accreditation Process
NAAC : Accreditation Process
 
Gene cloning
Gene cloningGene cloning
Gene cloning
 
Cancer
Cancer Cancer
Cancer
 
Multi purpose ID : A Digital Identity to 134 Crore Indians
Multi purpose ID : A Digital Identity to 134 Crore IndiansMulti purpose ID : A Digital Identity to 134 Crore Indians
Multi purpose ID : A Digital Identity to 134 Crore Indians
 
Techno Smart Card : Digital ID for Every Indian
Techno Smart Card : Digital ID for Every IndianTechno Smart Card : Digital ID for Every Indian
Techno Smart Card : Digital ID for Every Indian
 

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
 
WebAssembly is Key to Better LLM Performance
WebAssembly is Key to Better LLM PerformanceWebAssembly is Key to Better LLM Performance
WebAssembly is Key to Better LLM Performance
 
Behind the Scenes From the Manager's Chair: Decoding the Secrets of Successfu...
Behind the Scenes From the Manager's Chair: Decoding the Secrets of Successfu...Behind the Scenes From the Manager's Chair: Decoding the Secrets of Successfu...
Behind the Scenes From the Manager's Chair: Decoding the Secrets of Successfu...
 
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
1111 ChatGPT Prompts PDF Free Download - Prompts for ChatGPT
 
Where to Learn More About FDO _ Richard at FIDO Alliance.pdf
Where to Learn More About FDO _ Richard at FIDO Alliance.pdfWhere to Learn More About FDO _ Richard at FIDO Alliance.pdf
Where to Learn More About FDO _ Richard at FIDO Alliance.pdf
 
Overview of Hyperledger Foundation
Overview of Hyperledger FoundationOverview of Hyperledger Foundation
Overview of Hyperledger Foundation
 
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
Secure Zero Touch enabled Edge compute with Dell NativeEdge via FDO _ Brad at...
 
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
 
Salesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
Salesforce Adoption – Metrics, Methods, and Motivation, Antone KomSalesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
Salesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
 
How Red Hat Uses FDO in Device Lifecycle _ Costin and Vitaliy at Red Hat.pdf
How Red Hat Uses FDO in Device Lifecycle _ Costin and Vitaliy at Red Hat.pdfHow Red Hat Uses FDO in Device Lifecycle _ Costin and Vitaliy at Red Hat.pdf
How Red Hat Uses FDO in Device Lifecycle _ Costin and Vitaliy at Red Hat.pdf
 
ASRock Industrial FDO Solutions in Action for Industrial Edge AI _ Kenny at A...
ASRock Industrial FDO Solutions in Action for Industrial Edge AI _ Kenny at A...ASRock Industrial FDO Solutions in Action for Industrial Edge AI _ Kenny at A...
ASRock Industrial FDO Solutions in Action for Industrial Edge AI _ Kenny at A...
 
Intro in Product Management - Коротко про професію продакт менеджера
Intro in Product Management - Коротко про професію продакт менеджераIntro in Product Management - Коротко про професію продакт менеджера
Intro in Product Management - Коротко про професію продакт менеджера
 
Optimizing NoSQL Performance Through Observability
Optimizing NoSQL Performance Through ObservabilityOptimizing NoSQL Performance Through Observability
Optimizing NoSQL Performance Through Observability
 
Simplified FDO Manufacturing Flow with TPMs _ Liam at Infineon.pdf
Simplified FDO Manufacturing Flow with TPMs _ Liam at Infineon.pdfSimplified FDO Manufacturing Flow with TPMs _ Liam at Infineon.pdf
Simplified FDO Manufacturing Flow with TPMs _ Liam at Infineon.pdf
 
Easier, Faster, and More Powerful – Notes Document Properties Reimagined
Easier, Faster, and More Powerful – Notes Document Properties ReimaginedEasier, Faster, and More Powerful – Notes Document Properties Reimagined
Easier, Faster, and More Powerful – Notes Document Properties Reimagined
 
Syngulon - Selection technology May 2024.pdf
Syngulon - Selection technology May 2024.pdfSyngulon - Selection technology May 2024.pdf
Syngulon - Selection technology May 2024.pdf
 
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdfLinux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
Linux Foundation Edge _ Overview of FDO Software Components _ Randy at Intel.pdf
 
Designing for Hardware Accessibility at Comcast
Designing for Hardware Accessibility at ComcastDesigning for Hardware Accessibility at Comcast
Designing for Hardware Accessibility at Comcast
 
Enterprise Knowledge Graphs - Data Summit 2024
Enterprise Knowledge Graphs - Data Summit 2024Enterprise Knowledge Graphs - Data Summit 2024
Enterprise Knowledge Graphs - Data Summit 2024
 
AI revolution and Salesforce, Jiří Karpíšek
AI revolution and Salesforce, Jiří KarpíšekAI revolution and Salesforce, Jiří Karpíšek
AI revolution and Salesforce, Jiří Karpíšek
 

Fault Detection using Python

  • 1. Smart Manufacturing 4.0 RISHABH GARG BITS - PILANI I GOA IIT Delhi - AIA
  • 2. Fault Detection of Smart Lathe Machine Using Python Rishabh Garg BITS - PILANI K.K. Birla Goa Campus IIT Delhi - AIA Foundation For Smart Manufacturing .……………… Best Project Award Samarth Udyog 4.0 Indian Institute of Technology Delhi Department of Heavy Industry Ministry of HI & PE Automation Industry Association A Government of India Initiative Hosting Institute Government of India Industry Partner 1 3 A U G U S T 2 0 2 1
  • 3. Fault Detection of Smart Lathe Machine using Python Rishabh Garg BITS - PILANI | GOA IITD-AIA Foundation For Smart Manufacturing
  • 5. Project Background Problem Statement  Machines having large investments in capital also deploy high skilled labor for active operation and periodic maintenance. The challenge for companies is to protect this investment by ensuring high availability.  Unpreceded breakdowns in machinery and lack of skilled labor becomes a major challenge for manufacturing industries. Customer orders get delayed and company loses money  Solution: Deploy Machine Learning models and Real time dashboards and Predictive maintenance that constantly monitor the health of the machine and the associated tools. RISHABH GARG | BITS
  • 6. Project Background Physical Equipment on which it is implemented  No specific hardware required, apart from a regular PC.  GPUs can be used for faster training and results.  OS: Ubuntu – 18.04 LTS and Google Collaboratory (with GPUs) Unique challenge Development of accurate model despite high bias and limited dataset. RISHABH GARG | BITS
  • 7. LOREM IPSUM DOLOR SIT AMET, CONSECTETUER ADIPISCING ELIT. 12 Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Maecenas porttitor congue massa. Fusce posuere, magna sed pulvinar ultricies, purus lectus malesuada libero, sit amet commodo magna eros quis urna. 7 Solution RISHABH GARG | BITS
  • 8. Objective  Predict the instantaneous RMS value of vibrations of a Smart Lathe Machine using the traditional input parameters  Use the prediction for classification of the machine state into Normal and Faulty behavior.
  • 9. Objective  Display the prediction results through Streamlit Web App.  Implement the API endpoints using Flask.  Deployment to Heroku using Docker containers. RISHABH GARG | BITS
  • 10. Methodology Exploration of data  Data extraction from CSV files, Data preparation and filtering  Representation of data and its analysis (kurtosis, skewness etc. for appropriate model selection) Machine Learning Model Development Developed some sample machine learning models for checking accuracy and other metrics like sensitivity or positivity rate. Research was done on the models for hyper-parameter tuning etc. RISHABH GARG | BITS
  • 11. Methodology Flask Model was packed and API implementation was done using Flask Deployment through Docker containers Streamlit Necessary pathways for connection with dashboard using Streamlit Deployment to Heroku was done using CLI RISHABH GARG | BITS
  • 12. Implementation Data Extraction  Dataset provided from the FSM Lab consisted of raw data which was directly taken from the DAC device connected to the accelerometer.  For machine learning models to be built, first the required input parameters were identified.  RMS value of vibration was the best output parameter for predicting the machine state. RISHABH GARG | BITS
  • 13. Implementation Preparation of raw data  Using Python OS library, the vibration data stored in CSV files was aggregated and summarized in an Excel file.  60 One-Minute Experiments used as the training data for the linear regression model.  Redundant columns such as Dates, Timestamps, Serial numbers, Status etc. were removed and only numeric columns were retained.  RMS buffer values and Actual RMS values were used as output variables. RISHABH GARG | BITS
  • 14. Implementation Preparation of raw data  One-Minute experiments were used to predict the initial level of vibrations.  Full run experiments were used to determine the slope of RMS values with time.  RMS value were compared against the value provided by the line at a given time for classification. RISHABH GARG | BITS
  • 15. Implementation Preparation of raw data  Note: The data was very biased due to low occurrences of failure in the Lathe machine. This affected the accuracy of the model in the later stages. RISHABH GARG | BITS
  • 16. Implementation Training the model  Cubic SVM and Linear Regression models were compared for performance using MAE (Mean Absolute Error), RMSE (Root Mean Squared Error) metrics.  Linear Regression model was chosen out of the two due to higher dimensionality of data.  The threshold value of RMS vibration came out to be around 372.86 for the classifier model.  Logistic Regression was used to implement the classifier layer of model.  The predicted / actual RMS values were fed into the classifier for prediction of the machine state. RISHABH GARG | BITS
  • 17. Implementation Testing the model  After training both the models, the test dataset, generated using  Scikit learn library was fed the model and the predictions were stored.  The models were also evaluated based on a few metrics from the scikit learn library like MAE, R2, RMSE (for regression) and F1 score (for classification). RISHABH GARG | BITS
  • 18. Implementation Deployment  Heroku was used for the deployment of the final deliverable product which was connected to a Git repository for storage of the codebase.  A Docker file, Procfile and requirements.txt file was required for the appropriate stack selection by Heroku.  Streamlit was used for creation of the deployed frontend webapp and Flask was used to make the required API calls. RISHABH GARG | BITS
  • 21. Innovations  Real time monitoring of the input parameters, instantaneous and accurate prediction of RMS values of vibration data along with display of associated metrics was the biggest industrial level achievement of this product.  One can also select multiple models for prediction in the web app along with intuitive visualization. RISHABH GARG | BITS
  • 22. Innovations  This model, despite the associated bias, provides decent predictions due to proper choice of degree of SVR (here, behaves like Linear regressor). Tunable parameters have been added as sidebars in the webapp for customized accuracy as per the use case. RISHABH GARG | BITS
  • 23. Outcome Final deliverable product The following applications and repositories were made on the completion of the internship project for addressing the problem statement:  The app was deployed to Heroku as a frontend webapp: https://streamlitappml.herokuapp.com/  Source code for the Base models, Docker files and Flask: https://github.com/rishabhgargdps/IITD_ML RISHABH GARG | BITS
  • 24. Outcome Final deliverable product Source code for the Front end application: https://github.com/rishabhgargdps/IITD_streamlit Heroku app for customized models: https://streamlitml.herokuapp.com/ Heroku flask app that contains the API, and Docker containers: https://svmml2.herokuapp.com/ The codebase of the project was stored in Git Version control systems (here, GitHub). RISHABH GARG | BITS
  • 25. Scalability  This model can be used for prediction of RUL of any modern / traditional machine / tools if we can identify the most relevant input and output parameters pertaining to the quality of the product and the life of the machine.  The vibration data can be collected from the accelerometer attached to the tool holder which can be sent to the cloud for data storage using the Ethernet. There it can be processed and used for automated prediction.  Additional parameters like temperature, humidity of environment can be added for even better predictions. Thermo-hygrometers (DHT-11) can be used in the rooms for temperature and humidity monitoring. RISHABH GARG | BITS
  • 26. Scalability  One can also adjust the model for false positives / negatives as per need in the case of legacy hardware / biological equipment manufacturing.  The operator can be alerted using a web monitor service (e.g. Distil web monitor) running on the web app through an alarm sound, email notification etc.  Pre-trained neural nets or decision trees in a random forest can be used if we cannot acquire a large dataset and a similar use case has already been solved for a large dataset. RISHABH GARG | BITS