Niveau du stage : Master 2 ou 3ème année d’école d’ingénieur

Location: Labo LAME et LIFO, Université d'Orléans, France
Supervisors: Phi DO, Vincent NGUYEN
Duration: 6 months (approximately from March 1, 2025, to August 31, 2025)
Allowances: Legal internship allowances (around €600 per month, 4.35 euros per hour based on a 35-hour work week)
Deadline : 10 jan 2025
Keywords: PINN, Machine learning, RUL, time series data

Context:

In the field of predictive maintenance, estimating the Remaining Useful Life (RUL) of mechanical systems is important to keep things running smoothly, avoid downtime, and ensure safety. Traditional data-driven approaches often need a lot of labeled historical data, which isn't always available. Moreover, they don't work well in new situations. On the other hand, Physics-Informed Neural Networks (PINNs) integrate physical laws into neural network architectures, enabling them to work effectively even with limited data.

This internship focuses on studying state-of-the-art PINN methods for addressing challenges in RUL prediction. The goal is to deeply analyze these advanced techniques and understand their potential in solving real-world problems. The internship involves implementing and testing these methods on the CMAPSS dataset. Additionally, the methods can be integrated into a simple prototype of a Digital Twin framework.

Missions:

We aim to:

1. Investigate recent advancements in PINNs and their applications to RUL prediction.
2. Implement, improve and evaluate the performance of the existing methods on CMAPSS dataset
3. Integrating into a Digital Twins prototype.

Required Skills:

• Proficiency in deep learning and programming (Python).
• Background in machine learning, applied mathematics, or a related field.
• Familiarity with physics-based modeling and numerical simulation is a plus.
• Strong analytical and problem-solving skills.
• Ability to write clear and effective reports for regular updates and documentation of work.

Application:

Applications should be sent to duc-phi.do@univ-orleans.fr and vincent.nguyen@univ-orleans.fr with the subject "PINN RUL Prediction Internship Application" and include the following:

• CV
• Grades from previous academic years
• Name of two references

References:

[1] Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Maziar Raissi, Paris Perdikaris, George Em Karniadakis; Journal of Computational Physics 2019, Pages 686-707.
[2] Physics-informed machine learning: A comprehensive review on applications in anomaly detection and condition monitoring. Yuandi Wu, Brett Sicard, and Stephen Andrew Gadsden. 2024. Expert Syst. Appl. 255.
[3] Physics-Informed Neural Networks: Minimizing Residual Loss with Wide Networks and Effective Activations, Dashtbayaz, N. H., Farhani, G., Wang, B., & Ling, C. X. IJCAI 2024.
[4] Liao, Xinyuan, et al. "Remaining useful life with self-attention assisted physics-informed neural network." Advanced Engineering Informatics 58 (2023): 102195. 4.
[5] PINNsFormer: A Transformer-Based Framework For Physics-Informed Neural Networks. Zhao, Leo Zhiyuan and Ding, Xueying and Prakash, B Aditya. ICLR 2024.
[6] Cho, J., Nam, S., Yang, H., Yun, S. B., Hong, Y., & Park, E. (2024). Separable physics-informed neural networks. NeurIPS 2023.
[7] https://data.nasa.gov/Aerospace/CMAPSS-Jet-Engine-Simulated-Data/ff5v-kuh6/about_data