Title: Optimisation of replicated multi-response, non-linear quality characteristics with non-normal distribution through artificial neural networks

Authors: K. Orfanakos, G.J. Besseris

Addresses: Technological and Educational Institute of Athens, Agiou Spyridwnos 16, 12210 Aigaleo, Athens, Greece. ' Technological and Educational Institute of Piraeus, Petrou Ralli & Thivwn 250, 12244 Aigaleo, Athens, Greece; The University of West of Scotland, Paisley, UK

Abstract: Most related literature tackles quality characteristic(s) as single or simultaneous optimisation of multiple characteristics but through simple linear experimentation or mixture of linear orthogonal arrays. Due to their nature, characteristics demonstrate normal data distribution, analysed through traditional statistical methods or by computer aided solutions. This article tackles multiresponse quality characteristics through saturated non-linear OA experimentation that yields non-normal data distribution of a screening procedure in order to improve software design. Traditional analysis has limitations over such topics demanding a more sophisticated approach. Analysing such complex systems is achieved through artificial neural networks| function approximation ability which requires adequate data as training and testing exemplars that will produce a confident objective predicting system for the optimal control and level set. Experiment replication involves noise factors as being part of the screening process with a major role over the arrangements of the methodology proposed. E-mail spamming is used as a case study!

Keywords: multi-response quality characteristics; nonlinear quality characteristics; non-normal distribution; artificial neural networks; ANNs; expert systems; screening; software design; quality improvement; e-mail spamming; email spam.

DOI: 10.1504/IJQET.2010.035586

International Journal of Quality Engineering and Technology, 2010 Vol.1 No.4, pp.410 - 426

Published online: 30 Sep 2010 *

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