A Robustness Approach to Reliability

Abstract

Reliability of products is here regarded with respect to failure avoidance rather than probability of failure. To avoid failures,\ud we emphasize variation and suggest some powerful tools for handling failures due to variation. Thus, instead of technical\ud calculation of probabilities from data that usually are too weak for correct results, we emphasize the statistical thinking that\ud puts the designers focus on the critical product functions.\ud Making the design insensitive to unavoidable variation is called robust design and is handled by (i) identification and\ud classification of variation, (ii) design of experiments to find robust solutions, and (iii) statistically based estimations of proper\ud safety margins.\ud Extensions of the classical failure mode and effect analysis (FMEA) are presented. The first extension consists of identifying\ud failure modes caused by variation in the traditional bottom–up FMEA analysis. The second variation mode and effect analysis\ud (VMEA) is a top–down analysis, taking the product characteristics as a starting point and analyzing how sensitive these\ud characteristics are to variation.\ud In cases when there is sufficient detailed information of potential failure causes, the VMEA can be applied in its most\ud advanced mode, the probabilistic VMEA. Variation is then measured as statistical standard deviations, and sensitivities are\ud measured as partial derivatives. This method gives the opportunity to dimension tolerances and safety margins to avoid\ud failures caused by both unavoidable variation and lack of knowledge regarding failure processe

How to implement and apply robust design: insights from industrial practice

“…This definition supports the need for a more practical approach on RDM as research has shown several industrial challenges in applying RDM (Ida Gremyr et al 2003) and its surrounding statistical tools (Bergquist and Albing 2006). Novel research approaches on RDM (Ida Gremyr 2005b) and suggestions on how to operationalise it (Arvidsson et al 2006) have contributed to demystify RDM towards an extended audience and novel approaches (Johannesson et al 2013) are gradually complementing existing tools. Proposals have been made on how to integrate RDM in a generic product development process (Hasenkamp et al 2007); (Anna C Thornton 2004), some industrial insight on implementation is reported (Saitoh et al 2003a), (Saitoh et al 2003b), and research has searched for the practices ("what needs to be done") that joins the RDM-principles with the specific tools (Hasenkamp et al 2009 Arvidsson and Gremyr (2008) define the principles of RDM to be based on (i) awareness on variation, (ii) insensitivity to noise, (iii) application of various methods and (iv) continuous application.…”

Improvement in Product Development: Use of back-end data to support upstream efforts of Robust Design Methodology

“…Through that knowledge, settings of control factors that make the design of products insensitive to noise factors can be identified (Tsui, 1992). The sources which result in variations of product performance are traditionally categorized as: manufacturing imperfections (internal sources), environmental variables (external sources), and product deterioration (Johannesson et al, 2012). Manufacturing imperfections are seen in unit-to-unit variations of products due to manufacturing process variations.…”

Consideration of uncertainties in LCA for infrastructure using probabilistic methods

“…In these cases, the Variation Mode and Effect Analysis (VMEA) is a viable approach to analyse and quantify the uncertainty in the results of LCA models. VMEA is designed to systematically analyse the uncertainties affecting a model or a process and it has effectively applied to identify critical areas for reducing unwanted variation when developing a product or engineering concept (Chakhunashvili, Johansson, & Bergman, 2004;Johannesson et al, 2013). A performance function representing the outcome of the model is defined, generally expressed as dependent on a number of contributing uncertain factors which are treated as random variables.…”