Charting the State-of-the-Art in the Application of Convolutional Neural Networks to Quality Control in Industry 4.0 and Smart Manufacturing

This paper summarizes the theory behind state-of-the-art convolutional neural networks and investigates how they can applied to some of the challenges in Industry 4.0. Following an overview of key results in the literature, several major application areas are highlighted, with a focus on areas such as defect detection in production and anomaly detection for maintenance. It is concluded that convolutional neural networks are essential tools for solving many challenges in Industry 4.0.

I. Introduction

The human visual system is capable of recognizing and localizing objects within cluttered scenes with ease. For artificial systems, however, the goal of achieving the same level of competence has been plagued by challenges for decades, due to viewpoint-dependent object variability (i.e. in terms of scale and rotation), differences in lighting conditions and the open-endedness that comes with defining semantically relevant object categories.