Use of neural networks to predict roasting time and weight loss for beef joints

Abstract

A neural networks (NN) model was trained and validated using experimental data on roasting times and weight losses from beef joints. Mathematical and response surface (RS) models were also developed. Predicted results from NN and RS models were almost identical and better than the mathematical model. Using the trained NN and RS models, the effects of air temperature, dimension, weight of beef joint, its initial temperature on roasting time, and weight loss were investigated. An increase in air or initial beef temperature decreased roasting time but increased weight loss. For larger beef joints, both roasting time and weight loss increased significantly. Critical ratios of beef radius to length where roasting time and weight loss reached maximum values were found to be 0.45 using both NN and RS models for roasting time and 0.55 (NN model) or 0.6 (RS model) for weight loss. To improve productivity and reduce weight loss, small beef joints are recommended and beef joints with the critical ratios should be avoided.