Calculos de estabilidade e divisão de fases por meio de redes neurais artificiais / Phase splitting and stability calculations by means of artificial neural networks

AUTOR(ES)
DATA DE PUBLICAÇÃO

2006

RESUMO

Process simulation is a basic component of different Process Engineering activities such as On-line Optimization, Model Predictive Control, Identification, etc. The calculation of Phase Equilibrium appears as a fundamental task in any simulation of a separation process. However, the high computational time due to the iterative nature of this calculation makes it oft unsuitable for use with real time process analysis and synthesis strategies. The objective of this work is to develop a simple but accurate method to perform the phase equilibrium calculations required to the study of the behavior of complex systems. As such we mind those systems who present liquid-liquid and vapor-liquid-liquid phase equilibrium problems, such as systems with a heterogeneous azeotrope do. Given their inherent ability to learn and recognize non-linear and highly complex relationships, artificial neural networks (ANNs) appear to be well suited for such a task. Two chemical systems, the binary ethyl acetate ? water and the ternary ethanol ? ethyl acetate ? water were chosen; both systems present a miscibility gap and a heterogeneous azeotrope. The data sets used to train the ANNs were computed using the method of Pham &Doherty. Two kinds of neural networks were tried to solve the phase stability problem, namely the probabilistic neural networks (PNNs) and the perceptrons. In order to attain an acceptable precision perceptrons had to be trained with several hidden layers. Even though, PNNs got slightly better results than the perceptrons. Simple perceptrons were able to deliver the required precision when trained to predict the compositions of phases in equilibrium. Coupling the ANNs trained for phase stability with those trained for phase division a tool was obtained that can solve any phase equilibrium problem for the two chosen systems. Predictions made with the use of neural networks were faster than those made using the traditional methods, and delivered comparable precision

ASSUNTO(S)

phase diagrams equilibrio de fase phase equilibrium diagramas de fase azeotropes azeotropo

ACESSO AO ARTIGO

Documentos Relacionados