Design Of Experiments (Doe) And Regression

3 min read

Introduction Design of experiments is the useful techniques for controlling statistical quality control and also it’s a much complicated has a vital role in the field of optimization techniques.

DOE: The Jewel of Quality Engineering: These methods (DOE) all involve the activities of experimental planning, conducting experiments, and fitting models to the outputs.

Designof Experiments Methods: Five classes of experimentation and analysis methods are:

Method of DOE



Two-sample t-tests

Provide a relatively high level of evidence that a single level of a single factor causes a higher average response

Methods only address one factor-ate- time (OFAT). Compared with screening using fractional factorials, for comparable total costs the Type I and Type II errors are more likely.

Screening using Fractional Factorials


Provides an inexpensive way to determine which factors from a long list significantly affect system performance. Sometimes, users apply results to support final engineering design decisions

Compared with Response Surface Methods, the methods generate a relatively inaccurate prediction model. Compared with two-sample ttests, the level of evidence associated with significance claims is subjectively lower.

One-shot Response Surface



Create a relatively accurate prediction model and significance information, permitting identifying of interaction effects

Compared with factor screening methods, these methods require

Substantially larger numbers of experimental runs for a given number of factors.

Sequential Response Surface Methods


Generate a relatively accurate

Prediction model and may require fewer runs than one shot response surface methods.

The derived prediction model will, in

general, be less accurate than the one

from one-shot response surface

Methods if the method terminates without using all the runs.

Robust Design based

on Profit Maximization


Builds on RSM to directly maximize the sigma level in a cost-effective manner

addressing production noise

Complicated; may require substantial

experimental cost

Analysis of Variance

(ANOVA) followed by

multiple tests

Offers a standard approach for

analyzing significance of

factors and/or model terms

that addresses the multiplicity

of the tests

Compared with Lenth’s method and

Normal probability plots, the Type II errors are generally higher. This is

Only an analysis method that does not explain which data to collect.