Unified Uncertainty for Optimal Quality Decisions

I talked about the advantages of a unified uncertainty-based approach to quality in manufacturing earlier this year at LAMDAMAP. I have now gone into more detail on why a hybrid approach is required. In my presentation at ENBIS in Naples, I have given examples of how MSA and SPC can fail to properly quantify uncertainty resulting from systematic effects. Conversely, the approach of MSA and SPC is required to validate the theoretical models used in uncertainty evaluation.

Title: A Unified Uncertainty based Approach for Optimal Quality Decisions

Presented by: J E Muelaner

Conference: European Network for Business and Industrial Statistics 2017. Naples, Italy

Download the presentation: pdf link icon

Abstract:

There is no unified understanding of uncertainty in industrial processes and measurements, hampering data based decision making. Current methodologies include the Guide to the Expression of Uncertainty in Measurement (GUM), Measurement Systems Analysis (MSA) and Statistical Process Control (SPC). Although differences between methodologies are often simply linguistic in certain key respects there are fundamental differences. None of the methodologies are capable of providing a universal uncertainty framework for both processes and measurements. Additionally, methods currently aim to achieve fixed levels in metrics such as a ‘six-sigma’ process, a Cpk of 1.4 or a ‘Total GRR’ of 20%. With respect to business aims such as profitability, such arbitrary targets are not optimal for all processes. Although aiming for arbitrary targets based on an incomplete understanding of uncertainty has greatly improved quality, an optimized approach based on valid uncertainty statements will bring significantly greater improvements.
This presentation will first show where current methods are equivalent and where they are fundamentally different, leading to the development of a standardized vocabulary. It is then considered where fundamental gaps in each of the methodologies prevents them from providing a universal framework for industrial uncertainty. It is shown that new approaches are required to achieve such a framework. Finally it is demonstrated that a consistent approach to uncertainty, combined with Bayesian statistics, can enable arbitrary targets for process capability to be replaced by optimization of instrument and process selection and the specification of optimal conformance limits.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.