Date of Award

August 2020

Degree Type


Degree Name

Master of Science



First Advisor

Lingfeng Wang


To achieve efficient conversion and flexible control of electronic energy, insulated gate bipolar transistor (IGBT) power modules as the dominant power semiconductor devices are increasingly applied in many areas such as electric drives, hybrid electric vehicles, railways, and renewable energy systems. It is known that IGBTs are the most vulnerable components in power converter systems. To achieve high power density and high current capability, several IGBT chips are connected in parallel as a multi-chip IGBT module, which makes the power modules less reliable due to a more complex structure. The lowered reliability of IGBT modules will not only cause safety problems but also increase operation costs due to the failure of IGBT modules. Therefore, the reliability of IGBTs is important for the overall system, especially in high power applications. To improve the reliability of IGBT modules, this thesis proposes a new health state assessment model with a more sensitive precursor parameter for multi-chip IGBT module that allows for condition-based maintenance and replacement prior to complete failure.

Accurate health condition monitoring depends on the knowledge of failure mechanism and the selection of highly sensitive failure precursor. IGBT modules normally wear out and fail due to thermal cycling and operating environment. To enhance the understanding of the failure mechanism and the external characteristic performance of multi-chip IGBT modules, an electro-thermal finite element model (FEM) of a multi-chip IGBT module used in wind turbine converter systems was established with considerations for temperature dependence of material property, the thermal coupling effect between components, and the heat transfer process. The electro-thermal FEM accurately performed temperature distribution and the distribution electrical characteristic parameters during chip solder degradation. This study found an increased junction temperature, large change of temperature distribution, and more serious imbalanced current sharing during a single chip solder aging, thereby accelerating the aging of the whole IGBT module.

According to the change of thermal and electrical parameters with chip solder fatigue, the sensitivity of fatigue sensitive parameters (FSPs) was analyzed. The collector current of the aging chip showed the highest sensitivity with the chip solder degradation compared with the junction temperature, case temperature, and collector-emitter voltage. However, the current distribution of internal components remains inaccessible through direct measurements or visual inspection due to the package. As the relationship between the current and magnetic field has been studied and gradually applied in sensor technologies, magnetic flux density was proposed instead of collector current as a new precursor for health condition monitoring. Magnetic flux density distribution was extracted by an electro-thermal-magnetic FEM of the multi-chip IGBT module based on electromagnetic theory. Simulation results showed that magnetic flux density had even higher sensitivity than collector current with chip solder degradation. In addition, the magnetic flux density was only related with the current and was not influenced by temperature, which suggested good selectivity. Therefore, the magnetic flux density was selected as the precursor due to its better sensitivity, selectivity, and generality.

Finally, a health state assessment model based on backpropagation neural network (BPNN) was established according to the selected precursor. To localize and evaluate chip solder degradation, the health state of the IGBT module was determined by the magnetic flux density for each chip and the corresponding operating conduction current. BPNN featured good self-learning, self-adapting, robustness and generalization ability to deal with the nonlinear relationship between the four inputs and health state. Experimental results showed that the proposed model was accurate and effective. The health status of the IGBT modules was effectively recognized with an overall recognition rate of 99.8%. Therefore, the health state assessment model built in this thesis can accurately evaluate current health state of the IGBT module and support condition-based maintenance of the IGBT module.

Available for download on Friday, September 03, 2021