Fog-supported Low Latency Monitoring of System Disruptions in Industry 4.0: A Federated Learning Approach

In recent years, industrial alarm systems have undergone a considerable development, in terms of accuracy, complexity of their presentation and their management, and system reliability. However, alarm management systems suffer generally from the presence of poorly configured alerts, as well as numerous nuisance alarms (i.e., noise) or even alarm flooding events. This requires an immediate investment in finding methods and presentation tools to help operators understand the relationships between different alarms thus allowing an efficient maintenance management. To this end, we introduce in this paper a data mining approach to extract causal relationships between alarms to support maintenance operators & managers alarm during an alarm flood episode. We introduce the usage of Word2Vec algorithm (i.e., neural network for word embedding) as new clustering model for multidimensional data to identify groups of inherently linked alarms.
Such an approach should allow classification of alerts with a better understanding of the relationship between alarms, reduction of number of alarms and provides a solid basis for further causal analysis and identification of the root cause of the alarm. To demonstrate its merits, the proposed method is applied to alarms recorded during the observation of events in a post sorting center for the sorter unit and compared its performance with the state of the art clustering models.