Title: Implementing Mahalanobis-Taguchi system to improve casting quality in grey iron foundry

Authors: Vivek V. Khanzode, J. Maiti

Addresses: Department of Industrial Engineering and Management, IIT, Kharagpur, India. ' Department of Industrial Engineering and Management, IIT, Kharagpur, India

Abstract: Selection of influencing variables for multivariate process monitoring and control is a basic requisite for effective process improvement in manufacturing. Mahalanobis Distance is a useful distance measure in multivariate space and Mahalanobis-Taguchi System is a method of pattern recognition and data classification based on Mahalanobis Distance. In this study, Mahalanobis-Taguchi System (MTS) was used to identify variables influencing product quality. This study was carried out at a grey iron foundry. Data was collected over a period of three months and a total of 11 variables were considered for the study. The results of MTS were used to decide a variable monitoring scheme. Results were validated through simulated data. Significant improvements in capability indices were observed for the important quality variables. For example, maximum percentage improvement was observed for Carbon (75%), followed by Phosphorous (65%), Hardness (53%), Silicon (51%), and Sulphur (48%). While the improvement was reasonable for Carbon % and Silicon %, the absolute values after control were still less than acceptable limit of 1.33 and demanded further investigation and control.

Keywords: foundry process control; Mahalanobis-Taguchi System; MTS; healthy observations; unhealthy observations; orthogonal array; casting quality; grey iron foundry; pattern recognition; data classification; Taguchi methods; Mahalanobis distance; product quality; quality improvement; process monitoring.

DOI: 10.1504/IJPQM.2008.019760

International Journal of Productivity and Quality Management, 2008 Vol.3 No.4, pp.444 - 456

Published online: 29 Jul 2008 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article