Artificial neural networks modeling of non-fat yogurt texture properties: effect of process conditions and food composition

Abstract

Texture is one of the main characteristics involved in the acceptance of yogurt and must be monitored for the quality control of the product and for the adequate layout of the processing units. However, the determination of these properties requires expensive equipment, such as rotational rheometers, inaccessible to many industries. Thus, the modeling of artificial neural networks (ANNs) was applied to predict the texture properties of yogurt (output) based on changes in formulation and process conditions (inputs). Non-fat yogurts were produced with different centrifugation conditions and concentrations of protein and enzyme transglutaminase. Three models were developed: ANN-TPA to predict the properties obtained in the analysis of the texture profile (TPA), and ANN-TIX and ANN-VIS to predict thixotropy and viscosity, respectively. The enzymatic and protein variables had an impact of 46.3% and 53.6%, respectively, on the response of ANN-TPA. The shear rate had an impact of 83.7% and 86.0% on the thixotropy and apparent viscosity, respectively. The ANNs were able to predict responses with good precision (R2 >0.95) and low root mean square error (RMSE), showing their potential to be used as a tool to predict the properties of yogurt.