Potential applications of artificial neural networks to thermodynamics: vapor-liquid equilibrium predictions

The associative property of artificial neural networks (ANNs) and their inh erent ability to "learn" and "recognize" highly non-linear and complex rela tionships finds them ideally suited to a wide range of applications in chem ical engineering. Dynamic Modeling and Control of Chemical Process Systems and Fault Diagnosis are the two significant applications of ANNs that have been explored so far with success. This paper deals with the potential appl ications of ANNs in thermodynamics - particularly, the prediction:estimatio n of vapor-liquid equilibrium (VLE) data. The prediction of VLE data by con ventional thermodynamic methods is tedious and requires determination of "c onstants" which is arbitrary in many ways. Also, the use of conventional th ermodynamics for predicting VLE data for highly polar substances introduces a large number of inaccuracies. The possibility of applying ANNs for VLE d ata prediction/estimation has been explored using the back propagation algo rithm. The methane-ethane and ammonia-water systems have been studied and t he VLE predictions have been found to be accurate to within +/- 1%. Prelimi nary results confirm exciting possibilities of ANNs for applications to the rmodynamics of mixtures. Advantages and limitations of this application are also discussed. An heuristic approach to reduce the trial and error proces s for selecting the "optimum" net architecture is discussed. (C) 1999 Elsev ier Science Ltd. All rights reserved.