Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
Please note that all times are shown in the time zone of the conference. The current conference time is: 1st June 2024, 11:33:56am CEST
Efficient Multivariate Sensitivity Analysis for Electric Machines using Anisotropic Polynomial Chaos Expansions
Eric Emanuel Diehl1, Herbert De Gersem2, Dimitrios Loukrezis1,2
1Siemens AG, Germany; 2TU Darmstadt, Germany
This work suggests an efficient method based on anisotropic polynomial chaos ex- pansions for performing sensitivity analysis for multivariate model outputs. Generalised variance based (Sobol) sensitivity indices are used to quantify the sensitivity of the multivariate output to the model inputs. The suggested method is applied to an electric machine model which features vector-valued quantities of interest, e.g., the torque-speed characteristic. Comparisons against sensitivity analyses based on Monte Carlo sampling and isotropic polynomial chaos expansions reveal the significant accuracy and efficiency gains of the proposed method.
12:20pm - 12:40pm ID: 115 / Oral Session 3-3: 2 Abstract submission for online presentation Topics: Inverse problem Keywords: Deep learning, source-identification problem, magnetic field
A magnetostatic source-identification problem solved by means of deep learning methods
Sami Barmada1, Paolo Di Barba2, Nunzia Fontana1, Maria Evelina Mognaschi2, Mauro Tucci1
1DESTEC Department, University of Pisa, Italy; 2Dept. of Electrical, Computer and Biomedical Engineering,University of Pavia, Italy
In this work, a Deep Learning approach based on a Conditional Variational Autoencoder (CVAE) has been adopted for the solution of an inverse problems of magnetic field reconstruction knowing the field on a subdomain. Subsequently, starting from the CVAE outputs, the geometry of the field source can be identified. Two different techniques are used: a deep artificial neural network, fully connected, and a convolutional neural network. The proposed methods are applied to the TEAM 35 benchmark magnetostatic problem and a comparison between them is done.