Thermal properties estimation during thawing via real-time neural network learning Original Research Article

Predicting time duration of food thawing operations has a major importance in controlling the product safety and quality, as well as in optimizing and controlling the process in food industries. However this prediction, based on a model represented by a nonlinear distributed parameter system, depends essentially on a good knowledge of the thermal properties of the foodstuff. Instead of using classical differential scanning calorimetry or high order polynomial approximations, we propose in this paper to replace the estimation of these properties by real-time learning using simple neural networks. This network refinement is based on the Moving Horizon State Estimation and the reverse techniques. Experimental results were carried out during gelatin thawing, and are sufficiently good to now look forward to applying this method to real food, and to contribute further to the on-line control of thawing operations.