A neural tool for the prediction of the experimental dynamics of two-phase flows

Abstract

The dynamics exhibited by two-phase flows, which manifest themselves in a great variety of different flow patterns, are intrinsically complex due to the relevant number of degree of freedom, the nonlinear interaction of several phenomena and the uncertainty on the physical parameters. Therefore, an exhaustive mathematical modelling of two-phase flow dynamics is very difficult not only to assess and validate but also to extend and generalize to other applications. Nonetheless, a reliable model specifically oriented to the prediction of such dynamics, would represent an interesting step ahead towards the possibility of developing diagnostic or control tools for a variety of two-phase flow applications. The present study proposes the assessment of a short term prediction model derived from the experimental time series of the void fraction detected during an extensive experimental campaign for the characterization of vertical upward air-water two-phase flows under variable water and air superficial velocities, wsl and wsg. The identification strategy relies on the assessment of a NARMAX model (Nonlinear AutoRegressive Moving Average with eXogenous inputs) implemented in an approximated and generalized form by means of an optimized Multilayer Perceptron artificial neural network. Reported results show that a satisfactory agreement is reached between simulated and experimental data, showing that the model successfully predicts the time evolution of the void fraction dynamics. The application of a recursive feedback scheme to the model outputs allows to observe that satisfactory predictions can be obtained also for multiple steps ahead, though in the limits of a short-term predictability.

Hyperspectral Image Recognition Based on Artificial Neural Network

“…The analysis method quantifies the impact of input attributes on the output of the model, ranks the input attributes by the degree of impact, and eliminates those low impact ones. For dimensionality reduction, the NNSA was combined with BP neural network (BPNN) classifier, seeking to quantify the effect of each band on classification accuracy and determine the most impactful band combinations (Wang et al, 2016;Fichera et al, 2017).…”

Forming Quality Forecasting for Inner Threads Copper Tubes Based on Neural Network

“…(5) Update of the Weight and Threshold Value The error of output layer and hidden layer can be used to modify the weights and thresholds of the network (Fichera and Pagano, 2017).…”