Dynamic characteristics of a relief valve captured by a combination of fluid–solid coupling and feedforward neural network (FNN)

Considering multidisciplinary characteristics of thin plate vibration, fluid–solid coupling, and other aspects of a relief valve controlled by annular thin plates, a dynamic finite element (FE) model in view of fluid–solid coupling is firstly established for capturing relationships between dynamic characteristics of crucial indexes and partial working conditions. Secondly, the partial dataset of FE model under different conditions is statistically analyzed, and it will be utilized to train the feedforward neural network (FNN) model. The training process of FNN could be completed if results drawn from the FNN model are highly consistent with those of the FE model. Thirdly, dynamic characteristics under more conditions will be predicted through such a trained model, and dynamic behaviors from the FE model for same conditions of the FNN model are also obtained. Finally, comparing with results from the FE model, the maximum absolute error of steady-state displacement from the FNN is 0.0052 mm in an instance, thus verifying the rationality of this combined method. Consequently, such a combination of the FE model and the FNN model presents high accuracy and avoids repeated calculations of FE model with long times.