Development of a digital twin of electrical actuator based on a physical-informed machine learning model

The interest in digital twins has been demonstrated in several fields, such as energy, aeronautics, transport, medicine, etc. It consists of a virtual representation of a system aiming to connect the digital and physical systems. The advantages of this representation are multiple, notably the real-time collection of data for diagnostic and prognostic purposes and also the possibility of applying multitudes of tests that cannot be created/tested directly on the real system. However, for complex multi-physics systems with high dynamics and uncertainties, which is the case for energy conversion systems, developing a digital twin faces several challenges, including developing precise and fast models based on physics or data, and their implementation.

Indeed, the representation method influences the fidelity of the adopted solution. On one hand, the methods based on physics-based models, in particular derived from finite element simulation, can offer good accuracy but are characterized by high computational times that can be a drawback for real-time applications. On the other hand, data-driven methods are suitable with digital twins. However, they require adequate datasets, which are not easy to obtain and may not cover all the operating conditions of the system under study. Therefore, a hybrid representation combining physical knowledge and data-driven models associated with machine learning techniques could be relevant for developing efficient and high-fidelity digital twins.

Research subject work plan:

This PhD thesis aims to propose and implement methods for developing a digital twin for an electrical actuator. The model should reflect coupled electromagnetic and mechanical phenomena and also include uncertainties. This digital twin will be used to monitor the system's operation and health to extend the electrical motor's lifespan. An emphasis will be placed on integrating physical knowledge using artificial intelligence (AI). The PhD work will be developed with the following steps:

1. State of art

The first step of this work will be to elaborate on the state of the art on the following topics:

• Design of electrical machine,
• Datadriven modelling methods,
• Machine learning methods integrating physics,
• Classification techniques.

2. Contributions

The second step will be the development of the digital twin of the electrical actuator based on a physical-informed machine learning model. The model accuracy will be the key performance indicator for the different machine learning techniques that will be evaluated. The practical issues regarding the implementation of the digital twin will also be addressed. FPGA or GPU computing resources are potential candidates for implementing high-fidelity and accurate digital twins for real-time applications. Finally, the digital twin is expected to be an efficient benchmark to evaluate fault diagnosis methodologies for condition-based health monitoring.

References:

-Y. Wu, B. Sicard, S. A. Gadsden, ‘Physics-informed machine learning: A comprehensive review on applications in anomaly detection and condition monitoring.’ Expert Systems with Applications, Volume 255, Part C, 2024.

-Y. Fassi, V. Heiries, J. Boutet and S. Boisseau, ‘Toward Physics-Informed Machine-Learning-Based Predictive Maintenance for Power Converters—A Review,’ in IEEE Transactions on Power Electronics, vol. 39, no. 2, pp. 2692-2720, Feb. 2024.

-Singh, R.R., Bhatti, G.,Kalel, D., Vairavasundaram, I., Alsaif, F. ‘Building a Digital Twin Powered Intelligent Predictive Maintenance System for Industrial AC Machines.’ Machines 2023, 11, 796.

-Yigit A. Yucesan, Felipe A.C. Viana, ‘Hybrid physics-informed neural networks for main bearing fatigue prognosis with visual grease inspection.’ Computers in Industry, Volume 125, 2021.

-A. Sapena-Bano, F. Chinesta, R. Puche-Panadero, J. Martinez-Roman, M. Pineda-Sanchez, ‘Model reduction based on sparse identification techniques for induction machines: Towards the real-time and accuracy-guaranteed simulation of faulty induction machines.’ International Journal of Electrical Power & Energy Systems, Volume 125, 2021.

Située au Sud de Paris, au cœur d’un vaste territoire, composé de quatre départements franciliens et de nombreux sites stratégiques (Courtabœuf, Évry, Massy, Orsay, Saclay, Saint-Quentin-en-Yvelines, Vélizy et la vallée de la Bièvre), l’Université Paris-Saclay est un maillage unique de 5 facultés, 3 IUT, 5 grandes écoles, 2 universités membres associés et 7 organismes de recherche, réunis autour d’une ambition commune, celle d’associer leurs forces pour constituer un cluster scientifique de premier plan au service de la recherche, de la formation, de la réussite étudiante et de l’innovation, avec l’ambition de contribuer au développement d’une société plus juste et plus vivable. 1ère université française et dans le top 20 mondial, l’Université Paris-Saclay couvre les secteurs des sciences et ingénierie, des sciences de la vie et de la santé, et des sciences humaines et sociales.

Titulaire d'un master ou équivalent dans le domaine de l'énergie

Curieux et motivé