Methodological Foundations of AI for Anomaly Detection in Smart Manufacturing

This chapter presents a detailed insight into the use of Artificial Intelligence (AI) methods to detect anomalies in Smart Manufacturing. It initiates with a brief outline of the main challenges that accompany industrial data, such as high dimensionality, noise, imbalance, and non-stationarity, which subsequently lead to the suggestion of AI. In the next part, the chapter highlights different learning paradigms, including but not limited to supervised and unsupervised methods, semi-supervised, self-supervised, and reinforcement learning, and their advantages for manufacturing applications, which are discussed as well. Among the deep learning architectures today, the most attention is given to autoencoders, generative adversarial networks, and transformers, together with transfer learning for the model’s adaptability to different processes and domains. The chapter discusses not only algorithmic parts but also talks about the significance of explainability and human-in-the-loop refinement in the transparent, actionable, and trustworthy realization of anomaly detection. These combined features have made the techniques of human-centered and explainable anomaly detection stand out as the toughest part, while AI is perceived as an essential instrument that empowers durable, sustainable, and cooperative manufacturing at the time of Industry 5.0.