Automated Novelty Detection in Industrial Systems

Novelty detection is the identification of abnormal system behaviour, in which a model of normality is constructed, with deviations from the model identified as “abnormal”. Complex high-integrity systems typically operate normally for the majority of their service lives, and so examples of abnormal data may be rare in comparison to the amount of available normal data. Given the complexity of such systems, the number of possible failure modes is large, many of which may not be characterised sufficiently to construct a traditional multi-class classifier [22]. Thus, novelty detection is particularly suited to such cases, which allows previously-unseen or poorly-understood modes of failure to be correctly identified. Manufacturers of such high-integrity systems are changing the focus of their business such that they take on responsibility for provision of system maintenance [23]. Intelligent data analysis techniques are required to assess the “health” of system components, aiming to identify potential precursors of component failure in advance of actual system failure. This prognostic approach to condition monitoring is useful for types of fault that can be avoided if identified sufficiently early. In addition to providing early warning of critical failure, such techniques enable a “needs-based” approach to system maintenance, in contrast to the traditional dependence on maintenance scheduled at fixed intervals. Typically such warning systems require highly robust alarming mechanisms, with a minimal number of false positive activations, due to the cost involved of decommissioning and examining equipment following an alarm.