Martin Nachtsheim

Martin Nachtsheim

Auto-generated profile, based on publicly available information

Martin Nachtsheim

Department of Energy and Process Engineering, School of Engineering and Design, Technical University of Munich, Munich, Germany

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
13 February 2023 in IEEE Transactions on Transportation Electrification
Verified
Journal Article
Closed access
MNMartin Nachtsheim
Martin Nachtsheim
JEJohannes Ernst
Johannes Ernst
CEChristian Endisch
Christian Endisch
Ralph KennelRalph Kennel
Ralph Kennel
The recursive least squares algorithm for online parameter identification of induction machines has a high potential to serve as a basis for an innovative electric vehicle diagnosis concept. Commonly used for control parameter tuning, this approach is established in numerous industrial applications. However, in the automotive environment, special machine designs are used and highly dynamic operation takes place in a wide speed and load range. This results in fast transient parameter behavior, which is challenging in terms of online parameter identification. Therefore, the algorithm performance must be rated in our field of application with suitable dynamic test profiles. In this work, we compare several algorithm extensions to a novel recursive least squares algorithm with multiple variable forgetting factors for induction machines. The algorithms are analyzed regarding their handling of the associated transient parameter behavior. In addition, different identification model structures are considered to deal with the dynamic speed operation and the associated transient iron losses. Special attention is given to the real-time performance of the overall identification algorithms as this is a major requirement for implementation in automotive embedded systems. For validation, both, simulation and experimental results are presented, and associated configuration recommendations are provided.

15 June 2022
Verified
Conference Paper
Closed access
MNMartin Nachtsheim
Martin Nachtsheim
THThomas Hartmann
Thomas Hartmann
CEChristian Endisch
Christian Endisch
In the automotive area highly utilized electric traction machines are used to achieve the maximum efficiency with minimum space and weight. Their special technical design results in a strongly nonlinear inductive behavior. Detailed information on the inductive parameters and therefore the saturation behavior are important to gain highly efficient control of induction machines. Usually, these information are determined by complex standardized identification methods or specified commissioning runs. In this paper, a feed-forward multi-layer-perceptron is proposed to learn the steady state saturation behavior of an induction machine. For this purpose, characteristic equations are used to develop a physically inspired neural network. The training is proceeded offline based on standard testbench data sets of automotive induction machines. The model is optimized for implementation on embedded systems for real-time application with limited calculation capacity. Special consideration is given to the trade-off between accuracy and computational effort. To address this issue different configurations are implemented, varying the networks depth and the activation functions. The performances of the models are compared and discussed to show that no additional parameter acquisition is needed for the proposed approach.