Automated Identification, Classification, and Measurements of Pipe SurfaceDefects in Different Manufacturing Steps at Evraz

This research project focuses on Identification, Classification, and Measurement (ICM) of defects in different manufacturing steps at Evraz. Images/videos of the defects will be collected by a combination of the existing pipe inspection systems in use at Evraz and also the robotic system, equipped with a camera vision, to be completed in this project. Also, we will investigate about how economically beneficial the technology and its implementation would be. A literature review of previous efforts at other universities and/or industry will be first conducted to augment our pipe-defect data pool. Automated digital reporting on each pipe will be conducted in this project by using machine-learning-based techniques. The image/video databases will be utilized for training a Convolutional Neural Network (CNN). Field data will be divided into: (1) training, (2) validation, and (3) testing data sets. Training data will be utilized to train the network, validation data will be used to optimize the network architecture via hypothesis testing, and test data set will be used for testing the performance of the networks. This will lead to generating score cards for the pipes manufactured in Evraz.

Faculty Supervisor:

Mehran Mehrandezh;Christine Chan

Student:

Aidin Vahidmohammadi;Marzieh Zamani

Partner:

Evraz

Discipline:

Engineering

Sector:

University:

University of Regina

Program: