Development of an artificial neural network for the prediction of the thermodynamic property enthalpy in the NH3-H2O mixture

Abstract

There are many different methodologies for calculating the enthalpy thermodynamic property in the ammonia-water mixture, which is mainly used in the analysis of absorption refrigeration systems and power, so its prediction becomes essential not only for theoretical evaluations, also for the design of industrial equipment. In this work an alternative methodology, an artificial neural network (ARN) is approached. Two neural networks were designed: ARN A and ARN B. ARN A has three main input variables: Pressure (P), Temperature (T) and Ammonia Concentration in the mixture (x), to obtain the output variable: enthalpy. ARN B has as a particular case that the variable Temperature (T) is replaced by the phase in which the mixture is found (q); both networks were compared with experimental data reported in open literature and with the EESTM software. The two networks are capable of predicting the enthalpy of the Ammonia-Water mixture, ARN A with an acceptable prediction range between 100 kPa and 11,000 kPa, and ARN B from 5,000 kPa to 10,000 kPa.