Finite Control Set Model Predictive Control Combined with Online Junction Temperature Estimation for Reliability Enhancement of Voltage Source Inverters

As the efficiency of the power converter has significantly improved, its long-term reliable operation has become a critical factor in most power electronics applications. Among these factors, monitoring real-time temperature variations in the power converter during operation is a key priority for assessing reliability and predicting its lifetime. However, limited research has been conducted on the online monitoring of junction temperatures in power semiconductors using on-resistance measurements. Online temperature monitoring can help prevent individual device failure caused by transient thermal overstress, which could otherwise undermine the reliability of the entire converter. The power semiconductor junction temperature prediction method discussed in this paper still relies on the less commonly used direct measurement of on-resistance. Building on this approach, a combined method is proposed that integrates finite control set model predictive control (FCS-MPC) to minimize power loss and thermal stress in power semiconductors, thereby improving the overall reliability of the converter. Above all, acknowledging the variations in parameter distribution among identical power semiconductor devices, this method aims to estimate the online junction temperatures of all SiC power MOSFETs in a three-phase, two-level voltage source inverter. The simulation results demonstrate that the proposed method enables online estimation of the junction temperature of all SiC power MOSFETs in a three-phase inverter, and even allows closed-loop regulation of the junction temperature of the SiC power MOSFETs. Simultaneously, the reliability of the power MOSFET is notably enhanced due to the considerable reduction in junction temperature achieved by the proposed method. In addition, the single-vector MPC under finite set model predictive control, along with the double-vector and triple-vector approaches under continuous set model predictive control, are used for comparative analysis with the proposed method. After comparison, the excellence of the proposed method was finally verified through simulation.