Full text
| |
| Author(s): |
Kelly Moreira Gandra
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia de Alimentos |
| Defense date: | 2011-02-23 |
| Examining board members: |
Daniel Barrera Arellano;
Caroline Joy Steel;
Luiz Antonio Gioielli;
Renato Grimaldi;
Rodrigo Almeida Gonçalves
|
| Advisor: | Daniel Barrera Arellano |
| Abstract | |
The challenge for food industries to replace trans fats in various products lies in the development of formulations that yield fats with equivalent functionality and economic feasibility. Chemical interesterification has been used as the main alternative for obtaining zero trans plastic fats. Despite the technological evolution of raw material production processes, conventional methods used by food industries to formulate specialty fats are time-consuming and laborious and, in addition to calculations, many trial and error procedures are necessary. Neural networks are a field of computer science related to artificial intelligence, which has been used successfully in the area of oils and fats. Considering the difficulties faced by industries in the formulation stage of fats, the objective of this study was to apply the technique of artificial neural networks in the formulation of blends for zero trans biscuit fillings. Multilayer perceptron neural networks were constructed and trained using three raw materials: soybean oil and two interesterified fat bases. The neural network training phase was performed using as input variables the solid fat content and melting point of 62 examples of blends prepared with the three raw materials and, as output variables, the proportion of each raw material used in the different blends. The assessment of the learning capacity and efficiency of the neural networks in generalizing data was performed by requesting formulations of 16 blends used in training and 16 not used in training, respectively. The high performance of the neural networks to predict the solid fat content and melting point of blends formulated with the raw materials used for training was observed. To determine the range of application, formulations of fats for biscuit filling were requested to the network. Three formulations for each commercial fat used as standard were selected, all of which presented deviations greater than the solid fat content requested at temperatures of 10°C and 20°C. However, the solid fat content and the melting point determined experimentally for each formulation were very similar to those predicted. The fillings made with the formulations proposed by the network and commercial fats showed excellent thermal stability. The formulations proposed by the network, even though softer than the commercial fats, were able to maintain the structure of both filling and biscuit together without the expulsion of the filling. Neural networks can be considered a very valuable resource for the industry, as an alternative to conventional formulation procedures, as well as for the design and production of foods with low or zero trans isomer contents (AU) | |