Sensor Fusion for Real Time Quality Evaluation of Biscuit during Baking. Comparison between Bayesian and Fuzzy Approaches

摘要

In industrial baking, operators control the oven mainly by subjective evaluation of the color of the cookies. To standardize the quality of cookies, it should be interesting to replace the operator's subjective evaluation of color, by coupling a sensor and a data treatment tool like a classifier. With this aim, two different methods of classification were tested and compared: a fuzzy and a Bayesian method. Two kinds of results are presented: first, calibration and test points were used to establish the classifiers and second, a real time experiment was performed in which comparison is provided between operator classification and both fuzzy and Bayesian classifiers. The results for the two classifiers are good and coherent (73% of good classification for the two classifiers). Nevertheless, the fuzzy classifier is better adapted to this problem. Indeed, the notion of color manipulated by the operators is gradual and deterministic and not probabilistic. Furthermore, with such a fuzzy classifier, it is possible to propose a computer-aided algorithm that allows a predictive diagnosis of the evaluation of the classes of color.