Comparative analysis of robust design methods

Abstract

Robust parameter design is an engineering methodology intended as a cost effective approach to improve the quality of products, processes and systems. Control factors are those system parameters that can be easily controlled and manipulated. Noise factors are those system parameters that are difficult and/or costly to control and are presumed uncontrollable. Robust parameter design involves choosing optimal levels of the controllable factors in order to obtain a target or optimal response with minimal variation. Noise factors bring variability into the system, thus affecting the response. The aim is to properly choose the levels of control factors so that the process is robust or insensitive to the variation caused by noise factors. Robust parameter design methods are used to make systems more reliable and robust to incoming variations in environmental effects, manufacturing processes and customer usage patterns. However, robust design can become expensive, time consuming, and/or resource intensive. Thus research that makes robust design less resource intensive and requires less number of experimental runs is of great value. Robust design methodology can be expressed as multi-response optimization problem.

(cont.) The objective functions of the problem being: maximizing reliability and robustness of systems, minimizing the information and/or resources required for robust design methodology, and minimizing the number of experimental runs needed. This thesis discusses various noise factor strategies which aim to reduce number of experimental runs needed to improve quality of system. Compound Noise and Take-The-Best-Few Noise Factors Strategy are such noise factor strategies which reduce experimental effort needed to improve reliability of systems. Compound Noise is made by combing all the different noise factors together, irrespective of the number of noise factors. But such a noise strategy works only for the systems which show effect sparsity. To apply the Take-The-Best-Few Noise Factors Strategy most important noise factors in system's noise factor space are found. Noise factors having significant impact on system response variation are considered important. Once the important noise factors are identified, they are kept independent in the noise factor array. By selecting the few most important noise factors for a given system, run size of experiment is minimized.

(cont.) Take-The-Best-Few Noise Factors Strategy is very effective for all kinds of systems irrespective of their effect sparsity. Generally Take-The-Best-Few Noise Factors Strategy achieves nearly 80% of the possible improvement for all systems. This thesis also tries to find the influence of correlation and variance of induced noise on quality of system. For systems that do not contain any significant three-factor interactions correlation among noise factors can be neglected. Hence amount of information needed to improve the quality of systems is reduced.