A neural network-based evaluation method for the alignment state of air spring vibration isolation device

Abstract：Objectives The attitude control of air spring vibration isolation devices in marine propulsion systems was influenced by complex operating environments and changes in air spring pressure, leading to nonlinear and time-varying characteristics. Existing centering state monitoring models were usually based on rigid body assumptions and relied on precise system parameters, making it difficult to accurately describe the centering state under these conditions.【Method】 To address this, a BP neural network-based alignment state prediction model was established. The method of obtaining training and testing data under typical working conditions was formulated, and the data was processed with moving average noise reduction. The rules for adjusting the model's hyperparameters were summarized, and a systematic alignment state evaluation method based on the neural network model was proposed. 【Results】 Experimental studies were conducted on both small and large air spring isolation devices. The results show that the established neural network model can accurately predict the alignment state of the isolation device using only air spring pressure data, with strong generalizability across different types of devices. The prediction error is less than 0.5, and the alignment prediction accuracy reaches 96.29%. 【Conclusion】 The model does not rely on system parameters and performs well in predicting alignment states for both small and large devices. This research provides theoretical support for dynamic alignment state prediction and shaft alignment control of power equipment after startup.