Design of Experiments – Part 2

Design of Experiments – Part 2 article is in continuation of the first article on Design of experiments to help you understand better

Pranay Kumar Apr 24, 2019 2051 0

Design of Experiments

Design of Experiments – Part 2

Article is in continuation of the first article on Design of experiments to help you understand better.

Conducting an Experiment

DOE is commonly used in improve phase of six sigma (DMAIC) project. The first step in conducting experiment is screening of factors. If we want minimum cycle time then we first must identify/ screen factors that have significant impact on response variable. We even have some experimental designs available such as fractional factorial design or Taguchi design that we use to screen factors. In DMAIC methodology however, we commonly use hypothesis test to screen significant factors (X’s) out of all potential factors. Once we have significant factors identified then we design experiment and analyze two effect i.e. Main effect and Interaction Effect of the factor.

Main Effect

Let’s take one example; we just have two factors temperatures and pressure with two levels (i.e. Low and High). We conducted 2 experiments on following combination

Replication

One of the biggest limitations that OFAT experiment had was that we only conducted one experiment for each run. We must understand, that observation obtained from single experiment may not convey us true information about behaviour of response at that particular level of factor. What we may observe may be outcome of pure chance. For example, we conducted one experiment with temp and pressure both at its low level. Coincidentally, when we conducted this experiment, mixture was carrying raw material that was just half of its potential capacity. Ideally, we fill the mixture up to 90% of its capacity in regular production. But while conducting experiments with temperature and pressure at different levels, mixture might have been 90% filled. In such case you may get minimum cycle time with temp and pressure at its low level. But is that information valid? If we replicate experiment any such chance factors will be nullified. Minitab will also randomize the order for replicated runs.

To be continued….

*Please also read the DOE – Part 1 for complete article