Digital Twin & AI-based Control
Static load cycling tests play a key role in assessing the reliability of power semiconductors. To simulate real-world operating conditions, mission profiles are used to represent actual load scenarios. However, mapping these complex profiles on test benches is challenging, as suitable strategies for control and parameterization must be found. Artificial intelligence (AI) offers new possibilities in this context: With the help of learning and optimization methods, strategies can be developed that increase the accuracy and efficiency of test execution. The aim of this work is to develop and test AI-based methods that automatically translate mission profiles into suitable control strategies for static load cycle tests. The aim is to investigate methods that both ensure compliance with the mission profiles and enable efficient test bench operation.
Type of Thesis:
BA ❌ FA ✅ MA ✅
Relevant Experience:
- Basic knowledge of power electronics and semiconductor devices
- Interest in experimental work and data analysis
- Independent and structured approach to work
- Prior knowledge of microcontroller programming desirable
Contact:
The concept of the Digital Twin enables a seamless link between physical hardware and its virtual representation, providing new opportunities for system design, performance prediction, and reliability analysis. In this thesis, a 650 V embedded GaN half-bridge is modeled and investigated as a digital twin. The work includes the creation of an accurate electrical and thermal model, the validation of the simulation environment with experimental measurements, and the subsequent analysis of switching dynamics, losses, and reliability-relevant parameters. The digital twin will allow a deeper understanding of device behavior under different operating conditions and can serve as a foundation for further system optimization and real-time predictive control strategies.
Type of Thesis:
BA ❌ FA ✅ MA ✅
Relevant Experience:
- Basic knowledge of power electronics and semiconductor devices
- Independent and structured approach to work
- Prior knowledge of Matlab and Simulation tools preffered
Contact:
SiC power MOSFETs are widely adopted in automotive applications, such as three-phase traction inverters, for their high power density and efficiency. Accurate real-time determination of the junction temperature Tj ensures MOSFET reliability by maintaining operation within its safe operating area (SOA), preventing Tj from exceeding its maximum limit Tj,max. A widely used method for determining Tj is the use of lookup tables (LUTs). However, these require a lot of memory, which makes them difficult to use for real-time applications. For this reason, this paper will examine machine learning approaches such as a thermal neural network (TNN), which uses only TSEPs to determine Tj.
Type of Thesis:
BA ❌ FA ✅ MA ✅
Relevant Experience:
- Machine Learning
- Python
- MATLAB/Simulink
- C
- STM32
Contact:
Contact
Dominik Koch
M.Sc.Group Leader Power Electronics / Research Assistant