Quality Improvement by Using Inverse Gaussian Model in Robust Engineering

摘要

The concept of robust engineering (RE) which is based on the philosophy of Genichi Taguchi aims at providing industries with a cost effective methodology for enhancing their comptetive position in the global market. Since in most cases it is not possible to model the mathematical relationship between quality characteristic (QC), parameter designs and noise factors of situation under study, a proper statistical model in design of experiments (DOE) is proposed. However, the used statistical procedures in DOE are based on normality assumption of real data of QC or its transformed distribution. In many engineering cases, the data is highly skewed and therefore cannot be always removed by usual transformations; and even if it will be removed to a great extend, it may lead to inaccurate inferences in model parameters. Alternatively, the Inverse Gaussian family of distributions is flexible enough to provide a suitable model for these types of data. In this study, in dealing with such type of data, the concept of RE method is combined with Inverse-Gaussian (IG) model to reduce total deviations from target values of 17 quality characteristics in oil pump housings produced by Iranian diesel engine manufacturing (IDEM) company. As the distridution of data obtained from RE methodology follows the IG, the analysis without any data transformation (uncontrary in traditional RE procedure) is done straight forward through an IG model, and then its analysis is compared with customary analysis of RE method. This paper consists of four sections. The first section provides a brief description of problem. Section two gives a brief introduction to RE methodology. Section three devoted to introducing the proposed DOE model which is base upon inverse-Gaussian distribution. In section four, application of the two approaches to improve quality of produced oil pump housings in IDEM are considered and their relative results are obtained. And finally, in section five, the analysis results of application of the two models are compared.