Phase Stability Analysis Through Convolutional Neural Network

It is difficult to forecast the thermodynamic properties of multiphase systems because, in addition to equilibrium calculations, this requires find out how many and what kind of phases are present in the system (phase stability tests). The challenge is to create techniques that can determine whether a system is inside or outside of the two-phase envelope for a given overall composition, Pressure and Temperature, for instance (vapor and liquid phases in equilibrium, a single stable liquid phase or a single stable vapor phase). In this study, it is suggested to use artificial neural networks (ANNs) to solve the classification problem of phase stability. The inputs and output features needed to train the networks were obtained over the temperatures and pressures range in a constant composition that includes liquid-vapor equilibrium, liquid single phase and vapor single phase. Temperature and pressure are our inputs and the fluid phase at those conditions is our output. Therefore, the ANN needs to be able to select one of these three potential zones. Convolutional Neural Networks (CNN), a novel class of ANNs, were put to predict the phase of fluid. We examined several different CNN models with different conditions and finally presented the best network as the proposed network. According to the results, our suggested CNN were able to predict correctly the type of fluid phase in more than ۹۸% of the cases.

کلیدواژه ها:

Thermodynamic properties ، Phase stability ، Artificial neural networks ، Convolutional neural network ، CNN