Surface roughness (Ra) is widely used as an index of product quality, currently a strict technical requirement for many mechanical products. This paper introduces two models for Ra in peripheral milling processes: a) a statistical regression model based on cutting parameters, able to predict Ra in advance of the operation (off-line), and b) a novel artificial neural network (ANN) model based on cutting parameters and process variables that estimates Ra during the operation (on-line). Both models incorporate a soft sensor of the cutting tool wear condition (CTWC), which improves the Ra estimation. The soft sensor is represented by a hidden Markov model that integrates the Mel Frequency Cepstrum Coefficients of process signals. A principal component analysis (PCA) is used to reduce the complexity of the data. Results are validated by an extensive experimental program carried out on an industrial machining center.
