Fault Detection in Steel Surfaces Using Deep Learning Approaches

Abstract

There is sustainable growth in the industries in various developing nations. Quality maintenance of the product under development is an essential part of the product development process. Product quality affects the performance of the product. Various kinds of issues in the manufacturing can impact negatively on the product. Therefore, it is needed to make sure that the manufactured products are fault-free by establishing and employing such softwares that will ultimately bring some ease in the fault detection process. This paper aims to diagnose faults on steel surfaces by using convolutional neural networks and classification by making use of 5 different types of classifiers. They are Support Vector machines, Naive Bayes Classifier, Decision Tree, K-nearest Neighbors, and Random Forest. We have used 4 different types of models namely, Alexnet, InceptionV3, Resnet and VGG16. The testing accuracy was found to be maximum for the VGG16 model which was recorded to be 75.02%. Among the classifiers, the best accuracy was found out with Random Forest and Decision Tree classifiers to be 74.9% and 74.3% respectively. The defects are classified among the 4 categories of defects and are highlighted using image segmentation.