# Solving Industrial Problems with Machine Learning and image Processing, an application on Corrosion Estimation for failure prediction in O&G Pipelines.

sonam_AI

#### Description:

To know the Real Condition of Pipelines for Oil and Gas Pipelines is a resource intensive task. People have been doing Inspection for decades, All using NDT techniques for leakage and pilferage detection. But thanks to the huge advancements in computer vision and machine learning techniques the hundred percent safety of such industries is no more a pipe dream. When we talk about AI in industries it has been really misused as human replacing tool, but artificial intelligence is a tool for replacing zero failure and not zero employees.

So here is a a way of estimating Corrosion and including it with some weight in the failure prediction for a better understanding of the pipeline model. we have used techniques called GLCM and svm in this for understanding the data points of an image and its gradient with Entropy.

Outline of the Proposed talk:

1. Introduction Problem statement in the industry
2. Quick Introduction to Linear Regression and Logistic regression with one example each.
3. Understanding Support Vector Machines
4. Grey level Co occurrence matrix and its properties for image processing on corroded image
5. Extracting data from images using Scikit-Image, Image processing library
6. Application of SVM Algorithm using Sklearn on dataset collected
7. Visualizing data set for every example using matplotlib
8. application of SVM on non linear data set.
9. Other examples of machine learning on industry.

Content in the code of Github

a) Import Libraries

b) Upload the pics of high entropy less entropy and one test image

c) Extract image data and save it

d) Data Frame formation by stacking up values

e) Entering non corroded data en

j) Entropy and contrast with prediction value

f) Signing values to X as entropy contrast and Y as outputs (0/1)

g) Segregating points as corroded and non corroded only on the basis of output Y

h) Fit SVM in the data set

i) Getting the test image selecting one pixel and predicting its value for (0/1)

j) Use SVC.predict to get the value (0/1)

k) Get output in the form of corroded and non corroded

Pandas

#### Content URLs:

My Github account: The readme is very well Explained for the solution, Please Go through the Github text to understand it very well...

https://github.com/sonam-pankaj95/corrosion

My Slides for Presentation,

My quick video on the slides :

https://youtu.be/FymvOit5U2A

#### Speaker Info:

I am currently working as a Lead Software Engineer in IIT Madras incubated company. We work on data analysis of pipeline inspection and getting the customer know about huge risk or potential failure.

I also have one year of experience on working for autonomous vehicle, designing Control system and computer vision.

And was also appointed as a visiting faculty for Robotics and computer vision in PES University Bangalore.