RECENT ADVANCES IN ANOMALY AND DEFECT DETECTION USING MACHINE LEARNING IN SMART MANUFACTURING

Abstract

Smart manufacturing is a paradigm shift in conducting industrial processes, which is caused by the interdependence of
improved digital technologies and smart automation. This evolution is of utmost importance for anomaly and defect detection,
enabling a product with quality, efficiency, and predictive maintenance. This article reviews the key production techniques that
enable the creation of smart factory ecosystems: Cyber-Physical Systems, Big Data analytics, cloud computing, autonomous
robotics, and IIoT frameworks. It analyses the different levels of smart factory platforms, including mechanical design, process
control, IoT configuration, and communication infrastructure. Moreover, it evaluates machine learning methods for anomaly and
defect detection, including supervised, unsupervised, and semi-supervised approaches, and highlights their advantages and
disadvantages. It also reviews recent trends such as Digital Twin integration, Edge AI, privacy-preserving learning, transfer
learning, and explainable AI as indicators of new developments in industrial intelligence. Concomitantly, the paper identifies the
main limitations, including security risks, difficulties in system integration, and safety concerns in human–robot collaboration.
The present study provides unified views that can be utilised to enhance the reliability and efficiency of the smart manufacturing
sector.