基于物理信息神经网络的复杂管道流场加速求解方法OA

Complex piping systems exhibit coupling among their components.This makes it difficult to accurately predict the flow field using empirical formulas or a single theoretical method.Traditional computational fluid dynamics methods can resolve coupled flow fields in complex piping systems,but they require long computation time and high memory consumption.We investigate a data-driven method for flow field prediction in complex piping systems based on physics-informed neural networks.A training dataset is constructed from computational fluid dynamics simulation data,and a physics-informed neural network model is trained to accelerate the solution of flow fields in complex piping systems.The model directly embeds physical constraints into the loss function,so it can be trained with only a limited amount of training data.With spatial coordinates as input,the network outputs pressure and velocity at the corresponding locations.It reconstructs the flow field with good accuracy and enables rapid and accurate prediction.The results show that,with relative errors of 1.03%for the pressure field L2 and 13.28%for the velocity field L2,the trained physics-informed neural network model reduces the computational time by 94.52%in the inference stage,significantly improving the efficiency of flow-field solution in complex piping systems.