Smart Manufacturing IIoT Anomaly Dataset (SMIAD25)

This dataset represents a large-scale Industrial IoT (IIoT) multivariate time-series collection obtained from smart manufacturing environments operating under Industry 4.0 principles. The data span a continuous period from 17 January 2021 to 30 October 2025, recorded at 15-minute intervals, and capture machine-level operational behavior across multiple production lines, stations, and process stages. Each record corresponds to a unique (timestamp, machine_id) pair and reflects real operational dynamics observed in interconnected industrial systems. TThe acronym of dataset is Smart Manufacturing IIoT Anomaly Dataset (SMIAD25). The dataset contains heterogeneous feature groups commonly generated in modern manufacturing plants. These include mechanical and vibration-related measurements, acoustic indicators, electrical and drive characteristics, thermal and environmental readings, and process-level operational metrics. In addition, inspection- and vision-derived quality indicators are included to represent downstream quality assessment and surface inspection activities. Temporal evolution is characterized through change-based features, rolling statistics, and trend descriptors that reflect gradual degradation and long-term behavioral shifts. To account for inter-machine interactions, the dataset is accompanied by a machine relationship graph encoding proximity, material flow, and control dependencies among assets. This enables the study of relational effects and collective anomaly propagation in distributed manufacturing systems. The primary prediction target, anomaly_label, identifies abnormal versus normal operating conditions and exhibits strong class imbalance, consistent with real industrial fault occurrence patterns. Two auxiliary labels, anomaly_type and anomaly_severity, provide fault categorization and progressive intensity information, supporting advanced diagnostic and prognostic analysis. The dataset is suitable for temporal, graph-aware, and imbalance-robust learning approaches and supports realistic evaluation under deployment-oriented experimental settings.