Understanding precisely which X variables cause dramatic changes in NUD Y variables is important when optimizing the tolerance set points in any design or process. This knowledge is the basis for economical tolerance balancing and optimization. There is a difference between normal sensitivity, hyper-sensitivity and robustness of a function. When one or more control parameters (Xs) vary and the measured response Y changes dramatically as in an undesirable, non-continuous step function (affects adjustability) or worse - goes out of stable behavior, we have hyper-sensitivity. This is an internal-to-the-design parameter problem that indicates the foundational design concept has a flaw in its ability to be easily controlled and developed. This is not a robustness issue but more fundamental to the harmonious inner workings of the physics of the concept. Hyper-sensitive designs have to be corrected. Normally sensitive designs can be "lived with" usually by application of a control system.

 

ANOVA

% Contribution to Variation

Pareto for Hypersensitivity

The table below illustrates a tiered structure of % Knowledge-In-Hand to rate the risk a NUD Function or Characteristic is carrying due to the current level of data around Hyper-Sensitivity (ΔY/ΔX) of the measured response (Y) and specific control parameters (X_Control).

 

How to Assign % Knowledge-in-Hand for 5) Hypersensitivity
100%	I have clear ANOVA results. I can quantify F ratios and p values that describe any hypersensitivity between the Xs I intentionally varied within a DOE under controlled, nominal conditions. I know which Xs to include for Model Building DOEs in the future and the nature of their sensitivity.
75%	I have designed a special orthogonal array-based factorial experiment to exercise intentional small changes in my controllable Xs to study their hypersensitivity effect on the measured Ys that characterize critical functions.
50%	I have simulated the function and analytically identified sensitivities between controllable parameters and the output response(s) that need empirical investigation. I have a credible plan on how to measure and intentionally make very small changes in Xs to assess hypersensitivity on the measured Ys.
25%	I have identified the candidate Xs to intentionally change to assess hypersensitivity on measured Y values.
0%	I do not know which of my Xs have independent or interactive behaviors that cause extreme changes (hypersensitive reaction) in my measured Ys due to small changes in their level set points in prototypes or pre-production versions of the design.

 

The desensitization of NUD functional responses to unwanted sources of variation in our technologies, products and production processes receives almost universal desirability from technical professionals. When it's time to pay for that knowledge, many organizations pull their punches and rush past the important tasks of robustness optimization. CPD&M embraces robust design tasks as essential to designed experimentation that uncovers the important class of interactions between controllable Xs and undesirable and stressful noise factors (X_controls * X_noises). These are illustrated in the interaction plot, below.

Interaction Plot

This type of interactivity and sensitivity is due to 3 forms of unwanted variation called Noise:

1) Unit-to-unit/part-to-part/material-to-material variation,

2) External sources of mass and energy getting into the design from outside, and

3) Deterioration/wear-out/degradation sources that cause the function to change in Cpk-lowering forms.

Any CPD&M database that lacks the data around these interactions is incomplete and should be unacceptable to any Technical Leader or Gate Keeper.

The table below illustrates a tiered structure of % Knowledge-In-Hand to rate the risk a NUD Function or Characteristic is carrying due to the current level of data for Robustness of the measured response (Y) and specific control parameter and noise parameter interactions that leave the function insensitive to the noises.

There are 24 Critical Parameter Development & Management (CPD&M) Quick Guides being published in installments in this newsletter. Below are links to each of the prior newsletters with CPD&M Quick Guides: 

The CPD&M Quick Guide TOC (Nov 2013)

CPD&M QG1&2: Intro & Process (Jan 2014) 

QG3&4: Prioritize Req'ts & Design Guide (Feb 2014)

QG5&6: Functional Diagramming & Functions, Complexity & Risk (Mar 2014)

QG7&8: Fn's, Design Controls, DG O'view & I-O-C Diagrams (Apr 2014)
*Note: there was no May 2014 issue

QG9: Design Failure Modes & Effects Analysis (DFMEA) (June 2014)

QG10&11: Fishbone & Noise Diagrams (July 2014)

QG12&13: Base-line & Robust P Diagrams (Aug 2014)

QG14&15: Robust Design P Diagrams & Big 7 Metrics O'view (Sep 2014)

QG16&17: Measurability & Stability (Metrics 1&2 of 7) (Oct 2014)

QG18&19: Adjustability & Independence, Interactivity & Stat. Sig. (Nov 2014)