Engineers and Scientist are most often faced with two product development situations. One development situation is to find a parameter that will improve some performance characteristic to an acceptable or optimum value.
A second situation is to find less expensive alternative design, material, and method that will provide equivalent performance. When searching for improved or equivalent but cheaper design, the engineer typically runs some tests, observes some performance of the product, and makes a decision to use the design or reject the design. If the proper test strategies are employed, quality of the decision can be improved.
Not being aware of efficient and proper test strategies experimenter resorts to evaluate the effect of one parameter at a time. If one parameter does not work, another one is picked up for the test. In a desperate situation, the experimenter usually evaluates the effect of several situations sufficiently.
These methods have following limitations;
- If there is interaction then the same cannot be studie
- Data are not used efficiently
- Separation of main factor effect is not possible.
Full factorial experiments allow the experimenter to study two or more factors and their individual effects on the measured response.
Each factor can be studied at two or more levels. The experiment involves data collection at every possible combination of factors and levels, which allows estimation of effects for the main factors and interactions between factors.
Effect of main factors can be also studied in more detail, depending upon the number of levels chosen for a factor.
When two levels are used, a linear effect is measures.
As the number of levels increases, cubic and quadratic effects of the main factors can be studied.
Number of experimental Runs
Factorial experiment design requires, for most cases, a relatively large number of experimental runs. If ‘n’ number of factors is to be studied at two levels then total number of experiments required is 2.
The general formula is:-
Number of experimental runs
L = No. of Levels
F = No. of Factors
The number of repetitions for each of the experimental run should be between two to five. If the response is of attribute form, then the sample size should be large enough distinguish the output of one experimental run from other. If the data is attribute type and large sample is not feasible then nature of the defect is graded in the case of 1 to 10. Grade ‘1’ for the least severe defects, and ’10’ for hazardous or most severe defect.
The order of performing the tests of various trails should include some form of randomization. Randomization of the experiments protects experimenter from any unknown and uncontrolled factors that may vary during entire experiment, and any influence the result.
Randomization prevents bias of factors and interactions.
Randomization can take many forms, but three most commonly used approaches are
· Complete randomization
· Simple randomization
· Complete randomization
Complete randomization means that any trial has an equal chance of being selected for the first test. To determines which trial to run, draw a chit from chits having experimental run number written on it
If the experimental run has several repetitions then each trial should be randomly selected until al the trials have one test completed. Each trial is randomly selected in different order until all trials have two tests. The procedure is repeated till all repetitions of experimental runs are completed.
Download Excel Template
