|
|
|
| Deep Learning-Based Identification of Key Parameters for Landing Gear Buffer Systems |
|
|
|
|
Abstract To identify unmeasurable parameters in landing gear shock absorbers, such as the air polytropic index and orifice flow coefficient, a data-driven parameter identification method integrating a multi-scale convolutional neural network (MSCNN), bidirectional long short-term memory network (BiLSTM), and attention mechanism (AM) is proposed. Based on a nonlinear dynamic model of the oleo-pneumatic shock absorber, a mapping between drop-test response time series and system parameters is established. MSCNN is employed to extract multi-scale temporal features, while BiLSTM captures sequential dependencies. Attention mechanisms are further introduced in both feature and temporal dimensions to enhance the weighting of informative features and key time instants. Simulation results show that the mean relative error is below 1% under noise-free conditions. The robustness of the model was verified by adding measurement noise, reducing data dimensionality and decreasing the number of training samples, and the necessity of the design of each module in the network model was demonstrated through a series of ablation experiments. Compared with conventional temporal neural networks, the proposed method achieves superior accuracy and robustness, providing an effective approach for high-precision parameter identification of landing gear shock absorbers.
|
|
Received: 19 January 2026
Published: 28 October 2026
|
|
|
|
|
|
|
2026, Vol. 47 
