Estimation of Date Syrup Viscosity using Machine Vision and Artificial Neural Network

Science of rheology has numerous applications in various fields of the food industry such as process assessment, acceptance of products and sales. Fluid behavior changes during processing due to an adverse change in the consistency and due to the combined operations such as mixing, heating, cooling, etc. In this regard, viscosity is an important factor for quality assessment in most of the materials. To measure the viscosity, Viscometer devices are used which are directly in contact with the material. Working with these devices is time consuming, costly, under the influence of human factors and in some cases periodic calibration is required.
Date syrup was used as a viscous material in this study because it industrially is produced. An apparatus including a reservoir with an outlet orifice at the bottom was made to provide free flow of the liquid. Two sets of circular and rectangular orifices with different dimensions were used to investigate the effect of the orifice characteristics on the shape of the flow. Firstly, date syrup viscosity was measured by a conventional viscometer at 5 temperature levels and 6 concentration levels and behavior of the syrup were studied. Free flow of date syrup was photographed in the aforementioned temperatures and concentrations. On the other hand extracted features from the images were used as inputs to the neural network to give outputs as a fluid flow behavior index and consistency index. Measurement data were divided to three sets including training, validation and test sets whereas 70% of the data were used for training the neural networks, 15% as the validation set and 15% for testing the networks.
Results showed that similar to most of the liquids, viscosity of date syrup decreases when temperature increases. The experiments also revealed that the date syrup behavior is expressible with power law and can be determined using power equation. Date syrup has different behavior at different concentration levels. It changes from a pseudoplastic liquid to a Newtonian and then a dilatant liquid when concentration increases. Flow behavior index and consistency index corresponding to all three behavior of the fluid were determined. Results showed that the neural networks were able to accurately estimate the behavior and consistency indices with coefficient of correlations up to 0.99. Networks with three hidden layers were completely suitable for the estimation of the indices. These results revealed that in spite of different behavior of the liquid ranged from pseudoplastic to dilatant, the method was still able to determine the apparent viscosity of the fluid. Although the circular orifices were more efficient in determination of the indices than the rectangular orifices, there was not a significant difference between the uses of circular or rectangular orifices as well as no significant different between the orifices with different dimensions. The correlation between the actual and estimated values for fluid flow behavior index and consistency index was 0.98 whereas the mean square error of the validation sets was about 0.0138 which showed the accuracy of the method.
In this study a new method of viscosity determination was proposed. Machine vision was employed to estimate the viscosity based on the visual characteristics of the fluid free flow. Date syrup as a liquid with different rheological behaviors was used to assess the performance of the method. The strong correlation between the extracted features and fluid flow behavior index as well as a consistency index proved the reliability and accuracy of the method for viscosity estimation.