Prediction of food thermal process evaluation parameters using neural networks.

Area of Science:

• Food Science and Technology
• Chemical Engineering
• Computational Modeling

Background:

• Accurate prediction of thermal process evaluation parameters is crucial for ensuring food safety and product quality.
• Traditional methods for calculating parameters like g and f(h)/U can be time-consuming and may lack precision.
• Artificial neural networks (ANN) offer a powerful computational approach for modeling complex thermal processes.

Purpose of the Study:

• To develop and validate artificial neural networks (ANN) for predicting key thermal process evaluation parameters: g and f(h)/U.
• To assess the accuracy and reliability of ANN models in estimating sterilization values compared to traditional methods.
• To optimize ANN architecture and parameters for enhanced prediction accuracy in thermal processing.

Main Methods:

• Two distinct ANN models were designed: one for predicting 'g' and another for predicting 'f(h)/U'.
• Input variables included thermal destruction curve parameters (z, j(c)) and sterilization values (f(h)/U, g).
• Data from reported values were used for training and verification, with natural logarithm transformation applied to improve accuracy. 'Wardnets' with specific node configurations, learning rates, and momentum were employed.

Main Results:

• The ANN model for predicting 'g' achieved a mean relative error of 1.25% +/- 1.77% and a mean absolute error of 0.11 +/- 0.16 degrees F.
• The ANN model for predicting 'f(h)/U' demonstrated a mean relative error of 1.41% +/- 3.40% and a mean absolute error of 2.43 +/- 15.97.
• Process times calculated using ANN-derived 'g' values closely matched those from tabulated data, with a root mean square error (RMS) of 0.612 min.

Conclusions:

• ANN models provide a highly accurate and efficient method for predicting critical thermal process evaluation parameters (g and f(h)/U).
• The developed models can significantly aid in optimizing thermal processing and ensuring consistent product safety.
• The use of natural logarithm transformation and optimized 'Wardnet' architectures enhances the predictive capabilities of ANN in food sterilization.