Condition Monitoring of Discrete Power Devices: A Data-Driven Approach with Stress Quantification and Mold Temperature Sensing

Discrete power devices are used in a wide range of applications some of which would benefit from device condition monitoring (CM). This paper presents a data-driven method to online evaluate the health status of the discrete devices. It is based on the detection of mold temperature by only one thermocouple, offering low cost and non-intrusive features. The approach looks into five electrical stressors and uses existing sensors to track the load variation, enabling adaptation to both linear or nonlinear load conditions of the inverter system; no additional sampling of electrical signal is required. Under the tracked electrical stresses, a back-propagation neural network (BPNN) is employed to establish the correlation between mold temperature and aging status. The overall design and implementation process, including the hardware design, model training and CM integration, are demonstrated on a 7.5 kW T-type Neutral-point-clamped (TNPC) Uninterruptible Power Supply (UPS) inverter.