Effects of freezing rate and fluctuating storage temperature on quality of frozen bread dough and applications of artificial neural network for quality prediction

Abstract

This study investigated the effects of freezing rates, storage temperature and temperature fluctuations during frozen storage on frozen dough and bread quality. Quality changes were quantified by measuring dough weight loss, CO2 production rate, yeast viability, dough rheological properties, dough microstructure, dough water mobility (NMR), bread specific volume, bread crumb firmness and bread crumb characteristic. Two sets of quality kinetic data were measured at constant storage temperatures (-10oC, -15oC, -20oC and -25oC and -8oC, -13oC, -18oC and -23oC). Fluctuating storage regimes represented constant conditions (±0.1oC) good (±1.0oC), poor (±3.0oC) and very poor (±5oC) storage temperature control. Another regime mimiced temperature changes likely to be experienced in the cold chain (increases by up to 10oC). Both freezing rate and subsequent frozen storage had a significant deteriorative effect on all quality parameters. Slow freezing gave significantly better dough and bread quality than fast freezing. All quality parameters deteriorated further with increasing storage duration. Weight loss, CO2 production and bread specific volume were the most useful indicators of overall quality (low variability and correlated to sensory measures). Weight loss and decline in CO2 production during storage were adequately described by zero-order reaction kinetics. Weight loss during constant temperature storage was proportional to water vapor pressure consistent with the standard theory for evaporative weight loss from packaged foods. Large temperature fluctuations during frozen storage of ±3 to ±5oC and intermittent storage at higher temperatures resulted in significantly more rapid loss of dough and bread quality than storage at constant and/or lower temperatures. It is recommended that temperature variations should be less than ±3oC. An artificial neural network achieved a good fit (R2>0.98) between experimental and predicted data for weight loss prediction and seem to be promising for predicting bread specific volume loss (R2>0.7) but gave a poor fit (R2<0.5) for CO2 production and requires training data across the full range of conditions of interest.