Industrial Internet of Things (IIoT) rely on situation awareness (SA) to monitor operational states and detect anomalies using sensing data. However, existing SA frameworks rely on centralized processing and lacks to preserve abnormality information during hierarchical aggregation, making it difficult to localize the source of abnormal conditions. To address these challenges, a scalable multi-level situation awareness (MLSA) framework is proposed to enable bottom-up aggregation and top-down traceback with adaptive recalibration to localize abnormal conditions in hierarchical IIoT applications. In this work, first, a hybrid LSTM–1D CNN model is developed to identify operational states using time-series data at the edge level. Then, a metric-driven aggregation mechanism is presented to integrate anomaly descriptors and compute a composite situational score using an L 2 -norm at higher levels. The proposed method avoids transmitting raw data by propagating compact state descriptors, thereby reducing communication overhead. Finally, an adaptive threshold strategy is presented for selective traceback and local recalibration of abnormal branches. This allows IIoT applications to retain and reveal branch-level anomaly information during aggregation, resulting in a criticality propagation pattern in which abnormal conditions remain distinguishable. Experimental results across multiple datasets demonstrate that the framework preserves anomaly separability across hierarchical levels and enables more effective fault localization than centralized SA approaches.
Nasir et al. (2026) studied this question.