Title: Classifying tensile strength of HSLA steel: an investigation through neural networks using Mahalanobis Distance
Authors: Prasun Das, Shubhabrata Datta, Bidyut Kr. Bhattacharyay
Addresses: Indian Statistical Institute, SQC & OR Division, Kolkata 700108, India. ' School of Materials Science and Engineering, Bengal Engineering and Science University, Shibpur, Howrah 711103, India. ' Department of Mechanical Engineering, Bengal Engineering and Science University, Shibpur, Howrah 711103, India
Abstract: This paper addresses a comparative approach of classification of Thermomechanically Controlled Processed (TMCP) High Strength Low Alloy (HSLA) steels based on Mahalanobis-Taguchi System (MTS) principles and ensemble neural networks, including sensitivity analysis for variable selection. Later, a hybrid approach is developed, depending on the ability of Mahalanobis Distance (MD) in capturing the correlation structure of a multi-dimensional system, both for the Multi-Layered Perceptron (MLP) and for Radial Basis Function (RBF) networks. The results are found to be quite consistent in describing the role of input parameters for effective classification of such steel.
Keywords: HSLA; high strength low alloy steel; thermomechanical processing; classification; dimension reduction; MTS; Mahalanobis Taguchi system; artificial neural networks; ANNs; Taguchi methods; tensile strength; sensitivity analysis; variable selection.
International Journal of Mechatronics and Manufacturing Systems, 2010 Vol.3 No.1/2, pp.97 - 115
Available online: 02 Dec 2009 *Full-text access for editors Access for subscribers Purchase this article Comment on this article