A Bayesian network-based classifier for machining error prediction

Abstract

Machining error prediction is an important advance towards optimum machining. In this paper, we define the prediction problem as a classification task where process characteristics impacting on machining errors are attribute variables and error results are class variables. To deal with non-linearity and invisible correlations among process characteristics, the Bayesian network-based classifier learning from historic information is applied to extract hidden knowledge between process characteristics and resulting machining errors. Given an instance from the set of process characteristics describing, the posterior probability of each machining error result can be calculated using Bayes rules. An experiment for surface roughness (Ra) prediction on Al 7055 high-strength aluminum alloy in high speed cutting was presented, in which three Bayesian network-based classifiers, namely Naïve Bayesian Classifier, Tree-Augmented Network Classifier and General Bayesian Network Classifier, learnt from the experimental dataset simultaneity. Up to 84.6 % accuracy was achieved by GBNC which was selected as the surface roughness prediction model. Therefore, the Bayesian network-based classifier is a valid method for machining error prediction.