Using artificial neural networks to predict thermal conductivity of pear juice

Abstract

Thermal conductivity is an important property of juices in the prediction of heat- and mass-transfer coefficients and in the design of heat- and mass-transfer equipment for the fruit juice industry. An artificial neural network (ANN) was developed to predict thermal conductivity of pear juice. Temperature and concentration were input variables. Thermal conductivity of juices was outputs. The optimal ANN model consisted 2 hidden layers with 5 neurons in first hidden layer and the second one has only one neuron. The ANN model was able to predict thermal conductivity values which closely matched the experimental values by providing lowest mean square error (R2=0.999) compared to conventional and multivariable regression models. However this method also improves the problem of determining the hidden structure of the neural network layer by trial and error. It can be incorporated in heat transfer calculations during juices processing where temperature and concentration dependent thermal conductivity values are required.