Bayesian-inference-based neural networks for tool wear estimation

Abstract

This paper introduces the application of neural networks based on Bayesian inference, the automatic relevance determination algorithm for selecting relevant features and designing neural estimators for tool wear estimation in face-milling processes. Two types of neural networks are studied: Bayesian support vector machines for regression (BSVR) and Bayesian multilayer perceptrons (BMLP). Sixteen well-known features derived from pre-processing of the milling force signal are considered. The force signal samples are collected from 20 milling experiments under various machining conditions. The feature extraction and selection procedure is then applied to the sampled force signal. The feature selection results from the two neural networks are found to be quite similar. The average force has been proven to be the most relevant feature for tool wear estimation in both cases, among a set of six other features in each case, with each set differing by only one feature. The comparison among the generalization capabilities of the entire, selected, and rejected features shows that the selected features are relatively more relevant to tool wear processes in both cases. The comparison between the estimation results from the two neural networks using the corresponding relevant feature set shows that the BSVR method is more accurate in estimating flank wear than BMLP, but at the cost of a higher computing load.