Multi-scale spatiotemporal feature fusion MSCRA-GRU method for simultaneous faults diagnosis in aviation environmental control system

Fault diagnosis in aviation environmental control systems (ECS) faces significant challenges under simultaneous fault conditions due to high feature overlap, especially in critical components such as pressure regulated shutoff valves (PRSOV) operating under extreme conditions, including thermal shock, pressure transients, and vibrational fatigue. Conventional methods rely on static or single-fault hypotheses, which often struggle to distinguish overlapping fault characteristics, leading to misdiagnosis and delayed response. To overcome these limitations, this study proposes a novel multi-scale one-dimensional convolutional neural network with the residual module and attention mechanism, combined with the gated recurrent unit (MSCRA-GRU). The proposed method employs three parallel convolutional branches for hierarchical feature extraction, augmented by GRU to capture the temporal dynamics in PRSOV operational states, thereby enhancing spatiotemporal feature representation for both single and simultaneous fault scenarios. Validated on a multi-condition pressure dataset encompassing extreme operational scenarios, the MSCRA-GRU model achieves a diagnostic accuracy of 99.71% with a training time of only 77.65 seconds, demonstrating high computational efficiency. Comparison and validation with existing methods confirm that the proposed method outperforms traditional techniques in handling complex and overlapping dynamic signals, providing a practical solution with high accuracy and low latency for the intelligent maintenance of aerospace systems.