Design for Bread Baking Temperature Profiles using Neural Network Modeling Approach

Abstract

Various neural network models were developed to establish the relationship between tin temperature profiles and bread quality. The best model was composed of 6 input neurons, 6 first hidden layer neurons, 4 second hidden layer neurons and 4 output neurons with Log-sigmoid transfer functions. During verification, the correlation coefficient and mean square error were 0.9356 and 53.9229 respectively. To produce sandwich bread with various levels of crust color and weight loss, the best neural network model was used to design the tin temperature profiles for 4 baking zones. To obtain the same crust color and weight loss, the amount of increased tin temperatures for shorter baking time could be estimated. However the pattern of tin temperature profiles was not significantly changed. In contrast, the pattern of tin temperature profiles required for producing light crust color (L-values of 55, 65 and 55) and dark crust color (L-values of 50, 55 and 50) was significantly different. Therefore the neural network model presented the potential to assist industry with designing the baking profile to obtain the desire crust color pattern with shorter baking time and less weight loss.