Continuous (or Multiple) Target Problems

https://www.youtube.com/watch?v=Z9X3IhHytrQ

Another S/N ratio used is for problems where we do not have a single, static target value t for our performance, but we want to minimize the variation. This type of S/N ratio is appropriate for designs that have an offset adjustment. For example, we may produce a radio tuning circuit. We need to make radio circuits that do not vary off frequency with noise variations, such as heat, humidity, etc. The consumer can twist the dial to select the station; thus a bias does not matter. Once it is set, though, we want no variation. An S/N ratio appropriate for these type problems is

where p is the average across the noise,

This problem is very important i~ product development: we can usually move a bias with a change in d, but reducing variation is more difficult. This type of minimization is also known as a "two-model problem" in the statistics literature.

Minimization Problems

For problems which need p to be minimized, an S/N ratio of is suggested.

Maximization Problems

For problems that need p to be maximized, an SIN ratio of

is suggested

Selection of a Target Design

Having defined these S/N ratios, we select a value for each design variable d_i by using the value that maximizes the S/N ratio. That is, for each d_ih, evaluate:

where

For each d_i we evaluate this sum for each d_ih, and pick the value of d_ih that maximizes this sum. This approach applies a main-effect model, no interaction effects are considered. As discussed in chapter 18, often it is effective to plot the values of (19.10), to help understand the relative first order sensitivity of each variable. Two-way plots can also help visualize interactions.

Product Example

For the NerF Missilestorm ™ application, the chosen performance metric is travel distance of the missiles. The goal of the product is to maximize launch distance; thus, Equation 19.9 is selected as the appropriate signal-to-noise ratio.

Date: 2016-03-03; view: 789