Recent progress in AI-enabled compressor structural health monitoring

Zeji Li, Ligu Liu, Fa Zhong, Tianchi Lu, Ka-Chun Wong

Abstract: The rapid evolution of Artificial Intelligence (AI) and sensor technologies has transformed Structural Health Monitoring (SHM) from a reactive maintenance strategy into a predictive and intelligent system for compressors. This review synthesizes and evaluates the recent developments across compressor-relevant materials, sensing technologies, and AI algorithms that collectively enable this transformation. Relative to existing surveys, we explicitly focus on gas-turbine health management, rotating-machinery AI fault diagnosis, deep learning-based SHM, and SHM data fusion, and summarize the differentiating dimensions. The discussion begins with the relationships between material composition and dominant failure mechanisms, emphasizing on how titanium alloys, nickel superalloys, composites, and additively manufactured metals introduce distinct degradation patterns that shape monitoring requirements. Advances in sensing, including vibration analysis, acoustic emission, piezoelectric transducers, and fiber Bragg grating sensors, are critically analyzed in terms of their physical principles, sensitivity, and deployment limitations under harsh thermal and vibrational environments. Multi-sensor fusion and wireless architectures are highlighted as the key enablers for real-time distributed monitoring within energy-constrained systems. On the computational side, deep learning architectures (e.g., CNNs, autoencoders, and LSTMs) have improved the state-of-the-arts diagnostic performance and enabled data-driven RUL estimation, while transfer learning and explainable AI improve the adaptability and interpretability in complex operational contexts. To align with the PHM practice, this review distinguishes diagnostic metrics (e.g., accuracy/F1) from prognostic metrics that explicitly quantify the decision-relevant properties of RUL prediction (e.g., α–λ performance, prognostic horizon, and timeliness/asymmetric penalty for late predictions) and summarizes when each should be reported for compressor SHM. The emergence of Physics-Informed Neural Networks offers a promising avenue to embed physical knowledge within data-driven models, addressing both data scarcity and model transparency. In conclusion, future progress depends on multiple dimensions in experimental validation, cross-domain generalization, and the integration of physics-guided algorithms with advanced sensor networks to enable the robust and autonomous compressor monitoring system development. © 2026 The Author(s). Published by Elsevier B.V. on behalf of Faculty of Engineering, Alexandria University

External IDs:doi:10.1016/j.aej.2026.03.005