Uncertain data driven predictive maintenance: A cost-oriented implementation method on aircraft system

The increasing availability of on-board sensor data from complex systems, such as modern commercial aircraft, provides opportunities for developing data-driven health monitoring and predictive maintenance (PdM) methods. This paper proposes a health monitoring approach for commercial aircraft air conditioning systems, integrating unsupervised autoencoders with LSTM models to extract a health index (HI) and calculate the probability distribution of predicted performance parameters to represent system uncertainty. Additionally, a cost assessment model for predictive maintenance is developed to optimize maintenance decision thresholds based on the extracted health index. By simulating maintenance events for the aircraft air conditioning system within a predictive maintenance framework and applying the Gauss-LSTM-AE model to real data from a commercial aircraft fleet, this study assesses different health indicators from a cost perspective. The case study demonstrates that the proposed health monitoring method effectively identifies impending system faults. Moreover, the findings highlight that the integration of health monitoring with PdM decisions is significantly influenced by the health index and its decision threshold, which directly impacts system reliability and maintenance costs. This approach offers valuable insights into optimizing system safety by balancing predictive accuracy with economic constraints, providing a direction for improving reliability and efficiency in real-world maintenance operations.