Enhancing Modern Electric Powertrains with Advanced Data Analytics

Online thermal simulations are limited by computational constraints. Accurate modeling of the thermal behavior of electrical machines could reduce the need for excessive safety margins and unnecessary derating leading to higher performance of motors. This thesis implements model order reduction (MOR) methods to reduce the size of models while preserving accurate input-output relations. The goal was to create a sufficiently small model for use in online thermal monitoring applications with limited computational resources. Specifically the iterative rational Krylov algorithm (IRKA) and balanced truncation were implemented using the open source python library pyMOR. The system matrices of a finite element model (FEM) implemented in the Ansys Mechanical program were exported in an automatic manner using the pyMAPDL python interface. The subsequent MOR yielded good performance, creating highly accurate reduced order models (ROM) for all tested systems with ROMs consisting of no more than 25 states having negligible errors. The computational complexity of the implemented MOR was, however, great and for larger system sizes it could reach impractical levels; leading to the reduction process taking several days for sufficiently large systems. This was especially true for 3D models, which were found to be orders of magnitude more complex than equivalent 2D models. Specifically, balanced truncation was found to have reduction times exceeding 24 hours for 3D systems with more than 100,000 states, whereas IRKA exceeded such a limit around 300,000 states. For 2D models the computational complexity of both methods was lesser, such that a 767,000 state system could be reduced in less than one hour using either method. In general, the results imply the use of MOR may be a practical way to create small models for online applications. However, actual implementation and evaluation of the ROM in an online application was beyond the scope of the thesis.