Factories of the future are foreseen to evolve into smart factories with autonomous and adaptive manufacturing processes. However, the increasing complexity of the network of manufacturing processes is expected to complicate the rapid detection of process anomalies in real time. This paper proposes an architecture framework and method for the implementation of the Scalable On-line Anomaly Detection System (SOADS), which can detect process anomalies via real-time processing and analyze large amounts of process execution data in the context of autonomous and adaptive manufacturing processes. The design of this system architecture framework entailed the derivation of standard subsequence patterns using the PrefixSpan algorithm, a sequential pattern algorithm. The anomalies of the real-time event streams and derived subsequence patterns were scored using the Smith-Waterman algorithm, a sequence alignment algorithm. The excellence of the proposed system was verified by measuring the time for deriving subsequence patterns and by obtaining the anomaly scoring time from large event logs. The proposed system succeeded in large-scale data processing and analysis, one of the requirements for a smart factory, by using Apache Spark streaming and Apache Hbase, and is expected to become the basis of anomaly detection systems of smart factories.
Implementation of a general real-time visual anomaly detection system via soft computing
