Robust tool wear monitoring using radial basis function neural networks

Sunil Elanayar Veetil Thalakkurissy, Purdue University

Abstract

A robust monitoring scheme for tool wear is presented in this thesis. The nonlinear nature of most dynamic problems makes modeling and subsequent analysis difficult. A generic method for approximation of nonlinear systems is considered in this thesis. Radial basis functions provide a means to learn and model nonlinear systems based on a few sample runs. This property is used to develop models from experimental data. Subsequent usage of the models requires that a state estimation scheme be developed. An upperbound covariance method is presented to minimize divergence problems arising from modeling inaccuracies. Example applications of this method are presented in this thesis. A complete study of the use of these basis functions for representing tool wear evolution is investigated. Using prior experimental data, the basis functions are trained for subsequent use in monitoring. The effect of flank and crater wear on the cutting forces has been analytically modeled to minimize the number of experiments performed. The effect of flank wear on the indentation process has been studied to model force-wear relationships. The influence of crater wear on shear plane angle has been investigated to aid in prediction of force changes with crater wear. Predictions of forces for tools with flank wear has been shown to be more accurate than the crater wear case. By making use of the analytically determined output equations, a robust tool wear monitoring scheme has been implemented. Experimental results using carbide and ceramic inserts are considered. The estimation of flank wear is seen to be reasonably good, while crater wear estimates remain poor.

Degree

Ph.D.

Advisors

Shin, Purdue University.

Subject Area

Mechanical engineering|Industrial engineering

COinS