Comparison of Machine Learning Techniques of Damage Detection in Pipes Using Image Classification

Abstract

Pipes are the lifeline of every industry. Any damage to these pipes due to manufacturing defects or wear and tear can cause loss of resources and obstruction to normal activities. Detection of these damages at an earlier stage by human inspection methods is impossible. There is a need to find an effective method to detect pipe damage. This study aims at damage detection in pipe structures by using the image classification technique. A machine learning model is developed to detect the early-stage damages in the pipe. Machine learning (logistic regression, SVM, KNN, and random forest) and deep learning model (CNN) are used for developing the model. It is identified that the deep learning algorithm CNN has greater accuracy than any of the machine learning models, and hence, it could be used in real-time damage detection of pipes.