Identification of the degradation state for condition-based maintenance of insulated gate bipolar transistors: A self-organizing map approach

•Development of a SOM-based approach for IGBT fault detection•One SOM model is built for each component.•Setting the detection thresholds relying on few degradation trajectories•Method verified with IGBT experimental degradation data•Early detection of the degradation onset and of the maintenance intervention time

This paper presents an approach for the detection of the degradation onset and the identification of the degradation state of industrial components with inhomogeneous degradation behaviors due to the effects of multiple, possibly competing, degradation mechanisms and non-stationary operational and environmental conditions. The novelty of the approach is the use of dedicated Self-Organizing Maps (one for each component): each Self-Organizing Map is trained using data describing the component healthy behavior and a degradation indicator is defined by the distance between the test measurement and the Self-Organizing Map best matching unit. A case study regarding Insulated Gate Bipolar Transistors used in Fully Electrical Vehicles is considered. Data collected in experimental accelerated aging tests are used. The proposed approach is shown able to detect the initiation of the Insulated Gate Bipolar Transistors degradation process and to anticipate the component failure.