Statistical data fusion as diagnosis scheme applied to a kinematic chain

Abstract

In the modern industry framework, the application of condition monitoring schemes over electromechanical systems is being subjected to demanding requirements. Industrial electrical machinery implies the consideration of an electric motor, but also, gearboxes, shafts and couplings among others, resulting in complex kinematic chains. Such electromechanical configurations increase the risk of multiple faults coexistence and overlapping of corresponding effects in the considered physical magnitudes. Currently, the massive digitalization of industrial assets allows the investigation towards multiple monitoring strategies capable of emphasize deviations over the nominal system operation from different domains. In this regard, the proposed study presents the analysis of the diagnosis capabilities resulting from a high-dimensional statistical time-domain data fusion approach. First, multiple statistical time features are estimated from the available physical magnitudes, in this case, stator currents and vibrations. Second, two linear feature reduction techniques, such as principal components and linear discriminant analysis, are applied and compared. Third, a neural network based hierarchical structure is implemented for pattern recognition. The performance of the diagnosis scheme is compared with classical characteristic fault frequencies procedures. The study of the considered methodology is carried out over a kinematic chain driven by an induction motor under two fault scenarios, broken rotor bar and load unbalance.