Confidence in signal reconstruction by the Evolving Clustering Method

Monitoring the health conditions of equipment allows supplying advanced warning of their incipient failures; this can provide evidence useful to maintenance and replacement practices. However, uncertainties in the signal measurements and incompleteness in the representativeness of the measured data can overshadow the conclusions drawn from condition monitoring, and possibly lead the decision-maker to take wrong actions. In order to reduce the risk of wrong actions, confidence measures on the condition monitoring indications of the state of a component must be provided, so that the decision-maker can know to what degree he or she should trust such indications. As condition monitoring is usually structured in two modules performed in succession, one of reconstruction of the signal values in normal operating conditions and a following one of equipment health state diagnosis, it is reasonable to define confidence measures for the two processes individually, and then integrate the two into a single criterion for the whole condition monitoring. The research presented in this paper focuses on the definition of confidence measures for the signal reconstruction part of condition monitoring. The Evolving Clustering Method (ECM) is adopted to build the empirical model of signal reconstruction. Requirements for the reconstruction confidence are originally defined, and a single confidence measure is proposed to meet all the requirements identified. The confidence measure is analyzed with respect to two-dimensional artificial datasets and a real dataset concerning the Reactor Coolant Pump of a French Pressurized Water Reactor. The results obtained show that the proposed confidence measure meets all requirements and is more informative than the reconstruction error.