Explainable Anomaly Detection with Artificial Intelligence and Statistical Methods for Predictive Maintenance in Smart Manufacturing Systems: A Survey and Perspective

The COVID-19 epidemic has caused major problems in many parts of the world, but it has also led to the quick adoption of Smart Manufacturing (SM) technologies, especially as we move toward Industry 4.0 and 5.0. This rise shows how important it is to have reliable, understandable Predictive Maintenance (PdM) systems to cut down on downtime and make better use of resources. This survey provides a comprehensive analysis of recent advancements in Explainable Anomaly Detection (EAD), highlighting the synergistic integration of Artificial Intelligence (AI) methodologies—such as Variational Autoencoders (VAEs), Support Vector Data Descriptions (SVDDs), and Shapley Additive Explanations (SHAP)—with conventional statistical techniques, including Extreme Value Theory via Peaks-Over-Threshold (POT), Analysis of Variance (ANOVA), and causal inference frameworks, alongside Human-in-the-Loop (HITL) strategies to enhance transparency and human oversight. We create hybrid models that use machine learning to find anomalies with great accuracy. We also use statistical methods to help explain the results, such as feature attributions and root cause analysis (RCA) using the Five Whys, Ishikawa diagrams, and Fault Tree Analysis. A thorough real-world case study, employing the AI4I 2020 milling machine dataset, demonstrates this integration. We delineate forthcoming research trajectories, including innovations in causal discovery (e.g., PCMCI for time-series data), temporal explainable artificial intelligence methodologies, human-centric interfaces, and physics-informed digital twins, fostering interdisciplinary initiatives to improve robust, explicable, and human-centered predictive maintenance systems for resilient manufacturing ecosystems.