Complexity-based methodology for Fault Diagnosis: application on a centrifugal machine

In this paper a novel approach for the detection and the isolation of typical faults in oil refinery plants is presented. The proposed approach is based on a complexity-based methodology that allows monitoring a complex system by a holistic vision and provides a metric for the complexity measurements of the system. Applications for FDI of a centrifugal compressor are considered. Since this metric resulted sensitive to change in the operative conditions of the plant under study, for Fault Diagnosis purposes, a proper filtering procedure has been developed. System data relative to different operative conditions have been clustered and the information has been used to discriminate between variations of the complexity measure due to system failure from the one strictly related to changes in the operative conditions. The validity of the proposed method has been tested on real data concerning a fault occurred in a centrifugal compressor located in the Air Separation Unit (ASU) of a refinement plant.