Deep Learning Neural Networks for Heat-Flux Health Condition Monitoring Method of Multi-Device Power Electronics System

Semiconductor devices are usually used in parallel to increase the current rating, but uneven degradation is inevitable among the chips. Based on heat-flux heath condition monitoring (HHCM), one can diagnose the package degradation of individual chips using external measurements and a neural network (NN). In this paper several deep learning neural networks are analyzed in terms of training speed and accuracy. The paralleled devices are tested in steady state and transient conditions with variation of the operating point. Thermo-electrical measurement data are used for training and validating the NNs. The results show that a two-stage deep learning NN could increase the classification accuracy significantly and it is also possible to develop HHCM for field deployment in wind turbines.