Automated Faults Diagnosis Framework for Rotating Machines Under Combined Faults Scenarios

Ongoing technological advancements have made it possible to design very compact industrial rotating machines that offer significantly better performances with regards to conservation of energy and materials. Consequently, such compactness implies that several failure modes are becoming more intertwined, thereby increasing downtime associated with faults classification and diagnosis. Data fusion approaches such as coherent composite spectra (CCS) and poly coherent composite spectra (pCCS) have proven very useful for reducing overall faults classification times through the harmonisation of condition monitoring data from multiple locations into a single CCS or pCCS. This approach has made it possible for owners of physical industrial assets (PIA) to swiftly ascertain the condition of their assets without necessarily possessing significant expertise in rotating machine condition monitoring. As useful as the results of the techniques are, they have been solely applied for individualized or singular faults. However, rotating machines faults seldom occur in isolation, which makes it imperative to examine the proficiency of CCS and pCCS features under combined faults scenarios. This study considers 4 distinct cases of which 3 are singular (i.e. healthy baseline, bent shaft and shaft rub) and a combined bent-rub case. CCS data fusion was initially used to rationalize measured vibration data from all sensor locations, after which principal component analysis (PCA) and kernel principal component analysis (KPCA) are used to reduce output data dimensionality. Finally, the most representative features are then automatically classified using artificial neural network (ANN). The above steps are integrated to form a smart fault detection and diagnosis framework. Details of the approach and results obtained are described in the paper.