Speaker: Zhiguo Zeng, CentraleSupélec, Université Paris-Saclay

Inviter: 胡庆培

Title: Empowering predictive maintenance with AI and digital twins

Language: Chinese

Time & Venue: 2025.12.22 14:00-15:00 数学院南楼N402

Abstract: Predictive maintenance has become a key enabling technology for today’s industry. Recent advancement in deep learning has created great opportunities for data-driven predictive maintenance. However, the data-driven methods rely on large amounts of failure data with labels to train deep learning models, which are often too expensive and time-consuming to collect in practice. In this talk, we discuss some of our recent work aimed at addressing the limitations of data-driven approaches by digital twins and AI. In the first part of this talk, we present an open-source testbed for digital-twin-based predictive maintenance we developed in Centralesupélec. The platform comprises of an educational robot of 6-degree of freedom and the digital twin model of it. In the second part of this talk, we develop a digital twin supported deep learning framework to support data-driven fault diagnosis model. The developed fault diagnosis models are also able to diagnose component-level failures based on system-level condition-monitoring data. A central issue in using digital twin to generate simulation data for training data-driven fault diagnosis model is that the digital twin model is not guaranteed to represent sufficiently accurately the reality. To reduce the requirement on highly accurate digital twin model, a noise generation module is designed to simulate the discrepancy between the simulation and real data. This module significantly improves the performance of the diagnosis algorithm trained by an imperfect digital twin model. The proposed framework is evaluated on a real-world robot system. The results showed that the deep learning model trained by digital twins can diagnose the locations and modes of 9 faults/failure from 4 different motors.