To address the difficulty in accurately describing the states of alignment under nonlinear and time-varying conditions in existing monitoring models, a neural network-based method for evaluating the states of alignment is proposed.

A BP neural network-based prediction model is developed. The typical working conditions for acquiring training and testing data are defined, and the data is denoised using a moving average filter. The rules for adjusting the model's hyperparameters are summarized. Experimental studies are then carried out on both small and large air spring isolation devices.

The results demonstrate that the neural network model can accurately predict the states of isolation devices alignment using only the air spring pressure data. The model exhibits strong generalizability across different device types, with a prediction error of less than 0.5 and an alignment prediction accuracy of 96.29%.

The proposed model does not rely on system parameters and performs well in predicting the states of alignment for both small and large devices. The results of this study can provide theoretical support for the state prediction of alignment in a dynamic way and shaft alignment control of power equipment after startup.