Abstract: Direct monitoring of the pumping performance of liquid film seals is difficult to be put into effect. In order to test the performance parameters of the seals online, the monitoring method based on BP neural network is proposed. The pumpage and film thickness of liquid film seals under different pressures and revolving speeds are obtained through test firstly. The BP neural network is trained with experimental data. Output data of the network is gained by traversing input range method and the contour plots are drawn to be compared with the ones of measured data so as to assess the neural network generalization. Then, non-linear regression effects of five training functions are compared in three aspects involving generalization, accuracy and regression and the optimal BP neural network models are obtained. Finally, the monitoring effects of BP neural network are tested. And the results indicate that trainbr function is characterized by strong generalization and weak dependence on the number of hidden layer nodes. And BP neural network using trainbr function can satisfy the monitoring requirements for liquid film seals.

Key words: BP neural network, training function, liquid film seal, generalization