Condition Monitoring of Power Electronic Systems through Data Analysis of Measurement Signals and Control Output Variables


A major disadvantage of existing condition monitoring methods is the need for additional sensors and measuring equipment. In this work, this disadvantage is eliminated by completely avoiding additional hardware. Instead, software-based methods from the field of machine learning are used. Therefore, measurement signals and control output variables are utilized which are acquired and processed in any power electronic system for the purpose of converter control. The publication focuses on two main converter components: power semiconductors and DC link capacitor. For each component, the aging mechanisms that have been studied in the literature are explained. Based on the aging mechanisms, the degradation indicators are identified. Then, a converter model is built that allows the variation of degradation indicators in order to analyze their effects on the available data set. These findings form the basis for mathematical models, which detect future failure mechanisms of this type during converter operation. The test setup must offer the possibility of generating reproducible failure cases in various components with the aid of additional failure equipment. Finally, failure mechanisms are intentionally introduced at the test bench in order to validate the methodology of the developed approach.