Real-time defect detection and classification in robotic assembly lines: A machine learning framework
Fadi El Kalach, Mojtaba A. Farahani, Thorsten Wuest, Ramy Harik
Abstract
Manufacturing systems have witnessed a significant transformation with the introduction of Industry 4.0, introducing new capabilities with the emergence of new technologies. One such instance is the proliferation of sensors enabling the generation and acquisition of vast amounts of data, leading to advancements in Artificial Intelligence (AI) for manufacturing. One field profiting from this is that of Time Series Analytics (TSC) which includes forecasting and classification. TSC can be crucial for fault detection and diagnosis in manufacturing systems. However, there are still challenges in utilizing manufacturing datasets to train and deploy classification algorithms for real time classification. As such this paper aims to tackle these challenges by presenting a closed-loop framework for the testing and deployment process of TSC algorithms. This paper also details the feature selection and extraction process outlining specific criteria to be considered throughout. This is done by presenting a new manufacturing dataset acquired from a robotic assembly line and detailing the full process undergone in this study to train and deploy TSC algorithms on that manufacturing system. • This study tackles the challenges in utilizing manufacturing datasets to train and deploy classification algorithms for real time classification. • We present a closed-loop framework for the testing and deployment process of Time Series Classifications algorithms. • The feature selection and extraction process are discussed outlining specific criteria to be considered. • A robotic assembly line is used to showcase this framework.