Predicting the Remaining Useful Life (RUL) of industrial assets is a key aspect of intelligent maintenance systems, enabling proactive decision-making and cost-effective operations. However, developing accurate RUL prediction models presents several challenges, primarily due to the lack of large run-to-failure datasets and the complexity of degradation processes.
Challenges in RUL Prediction
Some of the key challenges in RUL prediction include:
- Data Scarcity: Real-world applications often lack representative run-to-failure data due to the rarity of failures in safety-critical systems.
- Domain Shifts: Changes in operating conditions can cause discrepancies between training and deployment data, affecting model performance.
- Degradation Complexity: The degradation process depends on multiple factors, including operational history and environmental conditions, making it difficult to model accurately.
Approaches to RUL Prediction
Our research explores various approaches to enhance RUL predictions, including:
- Hybrid Physics-Based and Data-Driven Models: Combining physics-based performance models with deep learning techniques to leverage domain knowledge and improve prediction accuracy [2,3].
- Uncertainty Quantification: Incorporating methods such as Bayesian neural networks and ensemble learning to estimate the confidence in RUL predictions and support risk-aware decision-making [4].
- Domain Knowledge-Informed Data Synthesis: Generating realistic degradation patterns using methods like scaled CutPaste and FaultPaste to enhance model robustness under extreme domain shift scenarios [5].
- Deep Learning-Based Health Indicators: Leveraging contrastive learning and autoencoder-based embeddings to develop interpretable health indicators that trend with system degradation [6].
- Operator-Based Approaches: Using Koopman operator theory to learn degradation dynamics and provide trendable representations for RUL estimation [7].
By integrating these approaches, our research aims to provide interpretable, scalable, and robust RUL prediction solutions that support reliable maintenance planning.
Lab Datasets:
N-CMAPSS is a new prognostics dataset [1].
[1] Arias Chao, M., Kulkarni, C., Goebel, K., & Fink, O. (2021). Aircraft engine run-to-failure dataset under real flight conditions for prognostics and diagnostics. Data, 6(1), 5.
Papers
[2] Ha, J. M., & Fink, O. (2023). Domain knowledge-informed synthetic fault sample generation with health data map for cross-domain planetary gearbox fault diagnosis. Mechanical Systems and Signal Processing, 202, 110680.
[3] Chao, M. A., Kulkarni, C., Goebel, K., & Fink, O. (2022). Fusing physics-based and deep learning models for prognostics. Reliability Engineering & System Safety, 217, 107961.
[4] Nemani, V., Biggio, L., Huan, X., Hu, Z., Fink, O., Tran, A., … & Hu, C. (2023). Uncertainty quantification in machine learning for engineering design and health prognostics: A tutorial. Mechanical Systems and Signal Processing, 205, 110796.
[5] Frusque, G., Nejjar, I., Nabavi, M., & Fink, O. (2024). Semisupervised Health Index Monitoring With Feature Generation and Fusion. IEEE Transactions on Reliability.
[6] Rombach, K., Michau, G., Bürzle, W., Koller, S., & Fink, O. (2024). Learning Informative Health Indicators Through Unsupervised Contrastive Learning. IEEE Transactions on Reliability.
[7] Garmaev, S., & Fink, O. (2024). Deep Koopman Operator-based degradation modelling. Reliability Engineering & System Safety, 251, 110351.