Disturbance Wave Velocity Model based on Physics-Guided Backpropagation Neural Network

Gas-liquid two-phase annular flow is an essential flow pattern in industries such as petroleum, chemical engineering, and nuclear energy. The interface wave's characteristics, especially wave velocity, considerably affect the heat and mass transfer, along with the resistance properties of two-phase flow. A dual-conductance ring sensor was developed to obtain the fluctuated liquid film thickness signal. The disturbance wave velocity was estimated by using the cross-correlation technique. A predicted model for disturbance wave velocity was proposed by combining Kumar interfacial shear stress model with a back propagation neural network (BPNN) with optimized network parameters. The predicted percentage errors (PEs) are within +/- 3.3% with a mean absolute percentage error (MAPE) of 1.20%. Compared with the model directly predicted by using BPNN, the predicted accuracy is improved, the PEs are within +/- 4.2% with MAPE of 1.55%, indicating the superiority of the physics-guided BP neural network in predicting accuracy, and is expected to extend the application scope.