Inductance Map Regression of Doubly-Excited Electrical Machines Considering Cross-Saturation

This article uses Gaussian process regression (GPR) to obtain the inductance map for doubly excited electrical machines under magnetic saturation, which is crucial for accurate control. The dc-excited field and ac-excited armature windings result in a cross-saturation between the winding inductances. Typically, finite element analysis (FEA) is used for this, but it is computationally expensive. In the evaluation of the inductance map, the GPR is selected due to its capability to efficiently minimize these computationally costly, but inevitably required simulations and/or measurements. This article presents the parameters affecting the inductance map and formulates the GPR method, trained with three different inductance datasets. Results show that the success rate of the inductance predictions while under magnetic saturation can be improved iteratively, from a 9.6% of maximum prediction error to 4.7%, by extending the training dataset within localized bounds.