Semantic Event Matching in Industrial Monitoring Systems with Incomplete Data

Abstract

Industrial monitoring platforms frequently rely on publish/subscribe mechanisms to disseminate sensor and equipment status information. In practice, event descriptions are often incomplete or inconsistent due to sensor failures and schema evolution. A semantic matching system was evaluated using 3.2 million event records collected from a manufacturing plant over 14 months. The dataset includes 1,740 distinct event types and 6,800 subscription rules. Ontology-guided similarity and rule-based inference were applied during matching. Compared with attribute-based filtering, recall increased from 0.71 to 0.86, while precision decreased marginally from 0.89 to 0.85. Average matching latency remained below 4.6 ms per event under peak loads of 12,000 events per second.