Effect of different basis functions on a radial basis function network in prediction of drill flank wear from motor current signals

Authors:
Saurabh Garg;Karali Patra;Surjya K. Pal;Debabrata Chakraborty
Affiliations:
Indian Institute of Technology, Department of Mechanical Engineering, 781 039, Guwahati, Assam, India;Indian Institute of Technology, Department of Mechanical Engineering, 721 302, Kharagpur, WB, India;Indian Institute of Technology, Department of Mechanical Engineering, 721 302, Kharagpur, WB, India;Indian Institute of Technology, Department of Mechanical Engineering, 781 039, Guwahati, Assam, India
Venue:
Soft Computing - A Fusion of Foundations, Methodologies and Applications
Year:
2008

Citing 0
Cited 1

Feature selection for high-dimensional machinery fault diagnosis data using multiple models and Radial Basis Function networks

Neurocomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

The Gaussian kernel has almost exclusively been used as the basis function of the cluster centers (hidden layer nodes) of a radial basis function network (RBFN) in most of its applications, especially in tool condition monitoring (TCM) problems. This study explores the possible usage of a set of five other basis functions in addition to the standard Gaussian function, in one such important TCM problem, i.e., prediction of drill flank wear. The analysis focuses on a comparative study of the wear prediction capabilities of the RBFN employing these six different basis functions for a wide range of the basis width parameter (wherever applicable) and changing the number of cluster centers in the hidden layer. This analysis is carried out following a series of experiments employing high speed steel (HSS) drills for drilling holes on mild steel workpieces, under different sets of cutting conditions (spindle speed, feed-rate and drill diameter) and noting the root mean square (RMS) value of spindle motor current as well as the average flank wear in each case. The results show that other basis functions can also match the performance of the Gaussian kernel, and depending upon the nature of application at hand and the requirements of time and space, the use of basis functions other than the Gaussian kernel may just prove advantageous.