MobilitY & reliability of Electrical chain Lab – MYEL

The initial phase of the research work under Axis 1 is based on two theses. The first thesis (Thesis 1) focuses on the development of classification methods using algorithms based on the analysis of time series obtained from sensors natively available on the motor/inverter drive train.
The second thesis (Thesis 2) focuses on the study of the contribution of magnetic field sensors to the early detection of short circuits between stator windings in permanent magnet synchronous machines.
The third thesis (Thesis 3) is being carried out in partnership with the EPEU laboratory at the University of Mons (Belgium). This thesis is based on the results of previous work. The main objective of the work is to estimate the lifetime of the electric motor winding.

Work completed Thesis 1.
- Study of the impact of using several voting systems in SeF-Rocket: multiple preference ordered
(SVEPMO), Highest Median (HMED), Possibility Theory (TP), Highest Mean (HMEAN).
(TP), Highest Mean (HMEAN).
- Implementation of a Threshold value to validate a vote or not. This method is
is based on the unanimity produced by the selected IR-PO (e.g. a vote may not be validated if
at least 18 of the 20 voters do not put the choice in their Top 5), otherwise PPV_MIX is chosen.
- Use of the StratifiedShuffleSplit in addition to the RepeatedStratifiedKFold.
search for the best IR-PO for large datasets.
- Creation of Hydra-SelFRocket, a method that concatenates the features produced by
Hydra and SelF-Rocket.
- Corrections, improvements and reformatting of the repository (SelF-Rocket Github) and updating of
diagrams and graphs.
- Set up a test environment on the MYEL compute & GPU server and retrieved the data required
the data required for the sensitivity analysis of the SelF-Rocket paper.

Work completed Thesis 2.
- The bibliographic research on the set of faults in a stator winding of a synchronous motor has been completed and has made it possible to develop analysis models that merge knowledge-based methods with advanced signal modelling techniques.
- The temporal signal analysis methods focus, firstly, on the associated Pearson correlation coefficient and, secondly, on CATCH22 and STL type methods.
- Pending the creation of an engine test bench adapted to the thesis problem, numerous measurement campaigns were carried out during the first year on a machine available at LSEE similar to our target machine.
- The engine under investigation was modelled in FLUX 2D software in order to map the radiated field for different fault configurations. A database is thus constructed and will serve as training data for our algorithms (machine learning).

Work completed Thesis 3
The thesis began in mid-September 2024.
The state of the art for this year is written as the first chapter of the thesis. It deals with the materials making up the insulation systems of rotating machines, the types of ageing of insulation systems (electrical stress under PWM inverters, etc.), the various ageing indicators, and predictive models for estimating the lifetime of electrical machines using artificial intelligence (AI).
- Started the first campaign of electrical and thermal ageing tests on specimens of two different wires in order to evaluate the indicators after each cycle.
- Produced a paper abstract for the SGE conference and defined the next steps

Thesis 1:
- Training on Pyleecan and, more generally, on the simulation of electric vehicle motor data.
electric vehicles.
- Initiate research on Near Real-Time default detection on microcontrollers and all related avenues
(e.g. Streaming Time Series Segmentation [1]).
- Configuration and preliminary tests on the SelF-Rocket NPU chip supplied (Rock 5b 16gb).
- Finalise corrections to the SelF-Rocket paper and resubmit the new version.

Thesis 2:
- Resolution of noise problems generated during measurements (sensor sensitivity)
- Validation of models already trained
- Developing models to define the degree of severity of faults.

Thesis 3:
- Continue the test campaigns until the insulation breaks down in order to build up a solid database.
- Carry out an in-depth analysis of the data for each indicator, to determine which one provides the best information on the two types of ageing (electrical and thermal).
- Develop a physics-based predictive model to estimate the lifetime of insulators.
- At the end of the year, carry out ageing tests on a complete machine winding in order to validate the predictive models on a more complex system.

- Mouad Talbaoui, Younes Azzoug, Sébastien Ramel, Remus Pusca, Eric Lefevre, et al.. Coefficient de corrélation et champ magnétique pour la détection de court-circuit entre spires dans les machines synchrones. 2024. ?hal-04693865?. JCGE (conférence nationale)
- Mouhamadou Mansour LO, Gildas Morvan, Mathieu Rossi, Fabrice Morganti, David Mercier. Évaluation de méthodes basées sur les noyaux de convolution aléatoires pour la détection de défauts mécaniques dans les machines
- arxiv.org/pdf/2409.01115

The common laboratory called "MYEL" aims to make the entire electric drive train more reliable by focusing on three main areas: making the components of this train more reliable, controlling them by means of appropriate control laws and predicting their failures. This LabCom is the initiative of the CRITTM2A (Centre de Recherche et d'Innovation Technique et Technologique Moteur et Acoustique Automobile) and the Laboratoire Systèmes Électrotechniques et Environnement (LSEE) of the University of Artois. The CRITTM2A is an independent test and expertise centre offering high value-added engineering, development and testing services in the fields of vibro-acoustics, batteries and electric and hybrid drive trains. The LSEE works on the design of electric machines that meet stringent requirements in terms of performance, durability and reliability; it also develops tools to diagnose their state of health. MYEL is part of a dynamic already initiated between the two partners with an ANR PRCE project launched at the end of 2021 in collaboration with two other companies. A first LabCom proposal was formulated in 2021; this version, which is more complete, is based on the remarks and advice given by the ANR evaluation committee.