Thesis Defense by Liwaa ABOU CHAKRA

The transition to electrification addresses major challenges, including reducing greenhouse gas emissions and decreasing dependence on fossil fuels. However, this shift presents new challenges for industry players, particularly regarding noise pollution generated by electric machines, which can affect users’ quality of life.

The design of these systems relies on an e-NVH approach that simultaneously integrates electromagnetic, vibrational, and acoustic constraints.

With this in mind, manufacturers rely on increasingly complex 3D numerical models from the early design phases onward, capable of precisely accounting for the machines’ geometric details and power supply.

These models specifically incorporate slotting, magnetic saturation, harmonics related to the pulse-width modulation of supply currents, as well as multiphysical phenomena and complex interactions between components.

Furthermore, accounting for uncertainties—such as manufacturing and assembly variations, rotor and stator eccentricities, and variations in material properties—all of which have been previously quantified experimentally—has become an essential step in the design process to achieve solutions that are both optimized and robust.

The main obstacle to this approach lies in the long computation times associated with the multiple finite element problem solutions required to explore the design space.

To overcome this limitation, this thesis proposes the development of reduced-order models dedicated to electromagnetic and dynamic problems, based in particular on Greedy Proper Orthogonal Decomposition and Double Modal Synthesis techniques, enabling the creation of numerical databases.

These models are then coupled with machine learning methods to conduct multiparametric and sensitivity analyses incorporating a large number of parameters derived from the various physical phenomena under consideration.

Mr. Franck MASSA, University Professor, LAMIH/UPHF, Thesis Co-Advisor
Mr. Stéphane CLENET, University Professor, L2EP/Arts et Métiers, Thesis Co-advisor
Mr. Thomas HENNERON, Associate Professor, L2EP/University of Lille, Thesis Co-advisor
Mr. Bertrand LALLEMAND, Associate Professor, LAMIH/UPHF, Thesis Co-supervisor
Ms. Pauline KERGUS, Research Fellow, LAPLACE, Examiner
Mr. Frédéric DRUESNES, Full Professor, ROBERVAL/UTC, Examiner
Ms. Emeline SADOULET-REBOUL, Associate Professor, FEMTO-ST/University of Bourgogne-Franche-Comté
Mr. Vincent LANFRANCHI, University Professor, ROBERVAL/UTC, Rapporteur

Electric Machines, Multiparametric e-NVH Analysis, Finite Element Method, Machine Learning, Greedy Proper Orthogonal Decomposition, Double Modal Synthesis.