Low Cost Defect Detection Using a Deep Convolutional Neural Network

Abstract

Starting with the late 1990s, more and more visual recognition tasks are being automated. The focus is around removing the human from the loop, thus increasing the overall process efficiency. Modern advancements in artificial intelligence allow us to start to automate tasks performed by specialized human beings, such as: visual inspection and defect detection. In this visual inspection process a deep convolutional neural network is used to fulfill the task of visual inspection of a bearing mounting process and classify it into 2 classes: good or faulty. This process is characteristic to the automotive sector. After training the CNN model the performances are evaluated by running the classification part of the algorithm on the validation dataset. No special training constraints are used and no special training method are used: everything is open source, as the goal of this research is to prove the viability for smaller companies to use AI in their everyday processes, in order to improve the overall business efficiency. The model can replace the human operator in the loop by having a mean accuracy of 90% and a mean processing time of 8 seconds. The overall cost of this application was well under the general market price for this kind of solutions. A path for further improvements is also demonstrated.