Design of Experiments – Part 1
This article on Design of Experiments shall help the reader understand the concept of Design of Experiments and how to use and design the experiment.
Design of Experiments – Part 1
We all as a human being right from our evolution to have always tried to control the natural phenomena around us. To control those phenomena around us it was imperative to understand the functioning of those processes and the most widely used method to understand those processes were experiments. We humans have always learned about the nature around us through experiments (mostly by wait and watch). We started cultivation and perhaps we tried every permutation and combination such as we added water to the soil, we added natural fertilizer to the soil, we tried different seeds, we plow the land and then cultivated and we observed the impact of these all factors on overall production.. Now we know which type of seeds used with which fertilizer under what condition will give us maximum yield. How could we control the yield? Obviously, we could control this by the knowledge that we obtained from the experiment. Changing all the factors (land/water/seed/fertilizer) at different levels (high/low, fertilizer A/ fertilizer B, etc.) and observing their impact on responses (yield) are experiments.
“Experiment is a systematic way of varying all the factors of interest and observing the impact of these all factors on the desired output”
We must understand that the experiments we carry out are more or less performed under controlled conditions and it may vary from simply “wait and watch experiment”. In wait and watch experiment we may not control the factors and so we may not get the best possible combination.
We all conduct several experiments every day. Ironically, we do not even realize that we are conducting an experiment. Often seen examples like restarting laptop when it’s not performing properly, taking the battery off and plug it back into the TV remote control when it’s not working, taking your mobile in open space when you struggle to find network coverage are nothing but experiment. Here we are changing levels of our factors (such as restarting laptop, changing battery, changing location of cell phone) to different permutation/ combination to find the best possible combination where we can get desired output.
OFAT experiment is a commonly used experiment in industry where people do not have knowledge of sophisticated technique like DOE (Design of Experiment). OFAT is an experiment where we just change one factor at a time keeping all other factors constant. Let’s take one example; we are working to reduce cycle time of a paint manufacturing process. We have identified temperature of ovens, pressure inside mixture and type of catalyst as a factor that may influence cycle time of the paint manufacturing process. In OFAT experiment we will keep level of all other factors (i.e. pressure and catalyst) and will keep changing the level of temperature (at all possible level) until we get a level of temperature where the cycle time is minimum. Once we identify optimum level of temperature then we will keep it constant (level of catalyst will also be kept constant) and we will keep changing pressure at different level until we get a level of pressure where the cycle time is minimum. Finally we will keep optimum level of temperature and pressure constant and will keep changing level of catalyst until we get optimum level of catalyst that causes least cycle time. This is the approach of OFAT, where we just change one factor at a time keeping all other factors constant.
This is widely used as the concept is easy to understand. But many experts will not prefer this technique because you may end up conducting large no: of experiments and what you conclude to be the optimum level of factors to produce least cycle time may not be exactly the true optimum level. As one combination in OFAT experiment is conducted only once, what you observe may depend largely on chance factor. One of the largest limitations that OFAT experiment has is, this does not give idea of interaction between two or more factors. We will be using concepts of DOE to carry out experiments in controlled way that will give us maximum information with minimum possible experiments.
We all know that a process is set of activities that require certain input and that produce certain output. But functioning of process is not always very simple. We always try to control output of the process around us for business purpose. We all want lesser cycle time, lesser rejection, higher productivity but to get optimum output we first must understand the complex process of interaction between various factors (inputs). While conducting experiment we manipulate these factors at various levels and then we observe (measure) the impact on response, this is how we try to uncover those complex interactions
Many of these factors cannot be controlled. So we allow them to vary, such factors are called noise variables. For example if you want to understand impact of temp., pressure and catalyst on cycle time then you are just controlling or manipulating these three factors. But you are not controlling many other factors such as humidity, skill/ experience of operators, raw material time, water content in raw material but even these factors may have certain impact on the output/ dependent variable. We call these variables are random variables or noise factors. And statistical calculation assumes that effect of these random factors cancels each other in long run.
* ******* The article is continued, please check Design of Experience -Part 2 for more**********************