Pipeline anomaly detection

PRESENTED BY:-
GAURAV BISWAS
BIT MESRA

 ABSTRACT
 INTRODUCTION
 Example and benefits
 LITERATURE REVIEW
 RESEARCH METHODOLOGY
 IMPLEMENTATION
 RESULT AND DISCUSSION
 CONCLUSIONs
 FUTURE SCOPE
 REFERENCEs

 The objective of this project is to locate and identify the
anomalies in an underground pipeline installation.
 Anomalies are identified and classified as structural damage
and Operational Maintenance also we have to trained the
Deep Learning model for Anomaly Detection and
Classification.
 For this we are using two techniques:-
(a) Convolutional Neural Network (CNN).
(b) Faster RCNN.
 Automation of manual assessment of hours of video.
 Reduce the human errors and reduce risk of liability issues.

 Pipeline Anomaly Detection is an automated Condition
Assessment System for pipeline networks. It reduces
manual efforts and time needed to review and code
video scans.
 It uses Artificial Intelligence (AI) and advanced neural
networks to identify, grade and score pipe anomalies.
 Videos of pipeline are generated by a camera-mounted
rover into the underground pipelines.
 It automates the process of identification of defects in
pipeline videos and generates a comprehensive
inspection report.

Fig: Camera-mounted rovers to record video of underground
sewer pipes.

Fig: Sample Anomalies and annotated images using the Pipeline
Anomaly Detection.

Saves Time.
Reduces Scanning Costs.
Provides High Accuracy.
Increases Availability of Experts.
Optimizes Maintenance Decisions.
Improves Customer Service.

Fig: Case Study of Pipeline Anomaly Detection.

AUTHOR YEAR STUDY FINDING
S. Safari
&
M.
Aliyari
2005 Detection and isolation
of interior defects
based on image
processing, Journal of
Pipeline Systems
Engineering and
Practice.
Principles regarding the learning
algorithm or deep architectures in
particular of those building blocks
unsupervised learning for single-
layer models and for pipeline
systems.
S. Kumar
& D.M.
Abraham
2017 Automated defect
classification in sewer
closed circuit using
CNN.
Investigated land area is labeled
into water and the classification is
compared to per-pixel works.
R.
Fenner
2009 Approaches to sewer
maintenance a review,
Urban Water pipeline
anomaly detection.
Using this approach we have to
reconstruct the frames from the
video and from that frames again
extracted the I- frames from it and
then display the result.

 Using Convolutional Neural Network (CNN) techniques for
implementing deep neural networks.
 Extracting Frames from the underground pipeline video and
then train those frames with this algorithm.
 Provide the underground pipeline frames dataset as an input.
 We have to use Optical Character Recognition (OCR) and
Pytesseract for extracting the text from the given images.
 Use OCR and Pytesseract for extracting text from image got
35% and 60% accuracy.

Also use Autoencoders for clearing the blur
images.
Extraction of I- Frames from the dataset using
different codes.
Detecting defects in the given video with a
camera-mounted rover and create a GUI
Application to get the output.
Got 90% accuracy using Faster RCNN.

For Extracting the text from the given pipeline video we
use:
 (a) Optical Character Recognition (OCR)
 (b) Pytesseract
(a) Optical Character Recognition (OCR):-
 It is the method of extracting text from the given image.
 It converts the image to gray scale, after that it smooth the image
and then it filters.
 Detect lines, words and characters.

(b) Tesseract:-
Python-tesseract is a tool used for OCR.
It is used an API to extract printed text from images.
It supports a wide variety of languages.
Tesseract was dependant on the multi-stage process
where we can differentiate steps like Word Finding,
Line Finding and Character Classification.

 Autoencoders:
 An autoencoder is a type of artificial neural network
used to learn efficient data in an unsupervised manner.
 The aim of an autoencoder is to learn a representation
for a set of data, typically for dimensionality reduction,
by training the network to ignore signal “noise”.

Fig: After performing Autoencoder this is the
outcome.

 If we want to extract just a single frame (I-Frame)
from the video into an image file we use I frame for
that.
 An I Frame (Intra coded frame) is a complete image
like a JPG or JPEG image file.
 The basic need is to compare the quality of the image
from the dataset and the actual I Frame image.

Fig: Actual Graphical User Interface (GUI) of Pipeline Anomaly
Detection.

 (A) Convolutional Neural Network:
 In deep learning, a convolutional neural network is a class of
deep neural networks, most commonly applied to analyzing
visual imagery.
 They are regularized versions of multilayer perceptrons. It
usually mean fully connected networks, that is, each neuron
in one layer is connected to all neurons in the next layer.
output

The Accuracy for Convolutional Neural Network (CNN) for this dataset is 70%
In this model we train our dataset into 50 epoch and finally get this
accuracy.

(B) FASTER RCNN:
Faster R-CNN with CNN features is the object
detection architecture and the pioneer approach
that applies deep models to object detection.
The function of this network is to generate
good features from the image.
The output of this network maintains the shape
and structure of the original image.

Fig: The architectures of Faster RCNN and Region Proposal Network
(RPN).

The Accuracy for Faster RCNN for this dataset is 90%.
In this model we train our dataset into 50 epoch and finally get this
accuracy.

 Using CNN and Faster RCNN to train our dataset for Pipeline
Anomaly Detection.
 A combination of CCTV, microwave sensor, neutron
and gamma ray, and hydro chemical sensors would be a
powerful tool.
 The most significant factor requiring water authorities to
undertake such exercises, however, will be the public
perception of sewer leakage.
 In our project the accuracy for CNN is near about 70% and
for Faster RCNN it is coming 90%.

Any deep learning application that we are using has big
scope in future. The function of output video frames needs to
be improved in the next stage of the research.
In addition, the output needs to be enhanced by including
more information (e.g., condition grade of the defects) to
realize the automation of the pipeline assessment.
Also the size of the training dataset is to cover as many
features as possible, which will lead to improving the
performance of the model.

[1] R. Fenner, 2002, Approaches to sewer maintenance a review Urban Water, New York, IEEE.
[2] K. Baah, B. Dubey, R. Harvey, 2009, A risk-based approach to sanitary sewer pipe asset management,
Germany, IEEE.
[3] M.D. Yang, T.C Su, 2014, Automated diagnosis of sewer pipe defects based on machine learning approaches,
Australia, IEEE, ICLR.
[4] Xinchen Yan, Jimei Yang, 2016, Conditional Image Generation from Visual Attributes, North Korea, IEEE,
ECCV
[5] O. Moselhi, 2004, Automated detection of surface defects in water and sewer pipes, Washington D.C, IEEE.
[6] S.K. Sinha, 2003, Computer vision techniques for automatic structural assessment of underground pipes,
India ECCV, IEEE.
[7] M.J. Anbari, 2017, Risk assessment model to prioritize sewer pipes inspection in wastewater collection
networks in Environ. Manag, Germany, IEEE ECCV .
[8] S.T. Ariaratnam, 2015, Financial outlay modeling for a local sewer rehabilitation strategy and Constr. Eng.
Manag, Russia, IEEE.

Pipeline anomaly detection

More Related Content

What's hot (20)

Similar to Pipeline anomaly detection (20)

More from GauravBiswas9 (11)

Recently uploaded (20)

Pipeline anomaly detection