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

Owing to the nonlinear deformation of thin annular plates, fluid-solid coupling behaviors and complex structural combinations, dynamic characteristics of shim valves could not be captured easily in a design stage in view of traditional methods. For obtaining dynamic behaviors accurately, a dynamic finite element model (DFEM) of a shim valve in view of fluid-solid coupling is developed firstly, thus presenting dynamic features under different pressure drops and structural parameters. Utilizing the partial data from the DFEM, the FNN model is trained, and it is further used to predict other dynamic characteristics under more conditions if high accuracy of such a FNN model is available. Finally, conditions including pressure drops and diverse structural parameters for predicted results could be introduced into a DFEM, and comparisons for dynamic results from two models will be achieved, further validating feasibility of such a FNN model. It is reflected through comparisons that dynamic behaviors from the FNN model are highly consistent with those of the DFEM whether the original or predicted data is considered, and relative error for the steady-state opening is less than 1 %. Therefore, such a combination achieves a good prediction for dynamic behaviors, and it could be used to design similar valves.