Artificial neural network model for prediction of cold spot temperature in retort sterilization of starch-based foods

作者：Llave Yvan Antonio; Hagiwara Tomoaki; Sakiyama Takaharu^*

来源：Journal of Food Engineering, 2012, 109(3): 553-560.

DOI：10.1016/j.jfoodeng.2011.10.024

• Artificial neural network
• Retort processing
• Starch dispersion
• Cold spot temperature
• F-0 value

An artificial neural network (ANN) model was developed for prediction of the cold spot temperature profile during retort processing using starch dispersion (STD) as a model food. STDs of different concentrations were prepared by mixing corn starch powder with distilled water at 90 degrees C for 30 min. Each of the partially gelatinized STDs thus prepared was filled in retort pouches and processed in a retort under various combinations of holding temperature, holding time, and rotational speed. Thermocouples were inserted into selected pouches one by one to monitor the cold spot temperature at regular intervals. The profiles of cold spot temperature together with retort temperature thus obtained were served to ANN modeling as training or validation data. Back-propagation network was chosen as the network model. Input variables for the model were current and past temperatures of the cold spot (T-n, Tn-1, and Tn-2) and current retort temperature theta(n) and current time t(n). Output was the temperature of the cold spot at the next time step Tn+1. A model with 2 hidden layers, which contained 11 and 15 nodes, respectively, was the best among the models tested. Using the model developed, prediction of a whole profile of the cold spot temperature was tested, starting from temperature data of the first three time steps with a whole profile of retort temperature monitored. The results showed very good performance of the model, relative errors for F-0 value prediction being less than 2%.

• 出版日期2012-4