Estudo do pré-resfriamento industrial de carcaças de aves usando métodos multivariados lineares e não-lineares

Abstract

Data extracted on an industrial process show a complex, linear or non-linear relationships between multiple variables. The use of multivariate methods is necessary to extract the information stored in data routinely collected by the industries. The aim of this work was to study the behavior of the chilling poultry carcasses in water-ice immersion chilling systems, based on data collected in a poultry slaughterhouse, using principal component analysis (PCA) and artificial neural networks (ANN). The PCA revealed a clear separation of samples from each line of chilling along of the first principal component (CP1). The topological and weight maps obtained from artificial neural networks (self-organizing maps – SOM) enabled us to analyze each process variable and its contribution to the segmentation of the samples. When comparing two lines of chilling it was possible to understand the main operational factors responsible for water absorption by carcass and their final temperature. The increase in the speed of slaughter, associated with high variation in carcass weight, demanded a higher thermal cooling load to the system for reducing the temperature of the carcass. The application of linear and nonlinear multivariate methods was effective in extracting information to control and understand the chilling process.