Modeling manufacturing processes using fuzzy regression with the detection of outliers

Empirical modeling, which involves various common techniques such as statistical regression, artificial neural networks and fuzzy logic modeling, is a popular approach to developing models for manufacturing processes. Among those techniques, statistical regression is the most popular one used to develop the explicit type of empirical models. However, if the experimental data and results contain a substantial degree of fuzziness, fuzzy regression is more appropriate for use in developing empirical models based on such data and results. In recent years, attempts have been made to use fuzzy regression to model manufacturing processes. However, it has been recognized that the existence of outliers can have a great effect on the prediction accuracy of a fuzzy regression model. This problem has not been well addressed in the previous studies on fuzzy regression. In this paper, an algorithm for detecting outliers based on Peters' fuzzy regression is proposed. The application of the algorithm to developing a fuzzy regression-based process model of the dispensing of fluid for IC chip encapsulation is described. Finally, the results of the validation of the models are discussed.