Applying Anomaly Detection to Fast Data in Industrial Processes

Big Data technologies and machine learning are about to revolutionise the industrial domain in different applications. Nowadays, industrial control systems are used to manage plant operations, to allow operators to control the activities of the entire plant, and to react to critical situations. A direct consequence of this, is that large amount of data is continuously generated, and it represents a significant information source. The analysis of historical data could then be applied in different scenarios, in order to support future decisions and prevent critical cases. In light of this, new prospectives for the development of analytic techniques are then available, but a major concern is the lack of appropriate platforms in the industrial domain. This project investigates the field of the anomaly detection techniques concerning the industrial processes, and it provides a dependable Big Data platform to be used as support to plant operators. The study has two important directions: first, the most appropriate architecture and technologies are pointed out; secondly, the most reliable algorithmic approaches are considered. The final outcome then consists of a complete fast, scalable and fault-tolerant platform, that offers anomaly detection services on historical and real-time data