Speaker
Description
Digitalization and automation in Industry 4.0 bring new opportunities to perform better maintenance in various industrial sectors, including the aviation industry. On-board sensors are used to monitor the health condition of aircraft components, and machine learning is used to predict the future of the health conditions. All these opportunities bring a new maintenance approach called predictive maintenance (PdM). PdM differs from the traditional paradigm called time-based maintenance, where maintenance activities are performed at fixed time intervals regardless of the actual health condition of individual components. Under PdM approach, we predict the remaining-useful-life (RUL) of individual components based on their current health condition data and plan maintenance activities in advance based on the predicted RUL. The goal is to prevent unexpected failures and unscheduled maintenance activities, improving the efficiency and reliability of aircraft.
Many studies have proposed various algorithms to predict the RUL of aircraft components. However, RUL prognostics have yet to be successfully integrated into maintenance planning. This presentation introduces a predictive aircraft maintenance framework to quantify the uncertainty of RUL predictions and plan predictive maintenance activities based on the uncertainty information. Convolutional neural networks and Monte Carlo dropouts are used to estimate the probability distributions of RUL of aircraft turbofan engines. Then, a deep reinforcement learning approach is used to suggest optimal maintenance decisions considering the probability distribution of the RUL. The proposed framework is illustrated using a digital twin model of turbofan engine maintenance.
In addition, the remaining challenges to implementing PdM in actual businesses are discussed. These challenges are identified based on literature, historical records of aircraft maintenance accidents, interviews with mechanics, and the presenter’s experience in interaction between academia and industry. Ultimately, the presentation aims to initiate an open discussion on the vision of predictive maintenance in Industry 4.0.
References
[1] Lee, J. & Mitici, M. Deep reinforcement learning for predictive aircraft maintenance using probabilistic Remaining-Useful-Life prognostics. Reliab Eng Syst Safe 230, 108908 (2023).
[2] Lee, J., Mitici, M., Blom, H. A. P., Bieber, P. & Freeman, F. Analyzing Emerging Challenges for Data-Driven Predictive Aircraft Maintenance Using Agent-Based Modeling and Hazard Identification. Aerospace 10, 186 (2023).
| Keywords | predictive maintenance, aircraft maintenance, deep reinforcement learning, prognostics health monitoring |
|---|
